Ok I made a mistake.
The game harness was doing a lot of heavy lifting for both models by preparing optimal options for Jev/Haiku to pick from, given a game state.
If I change the harness to provide only game rules and game state, and only allow Jev to decide the control buttons (left/right/rotate/drop), it pretty much becomes useless.
引用@tdinh_meTetris battle: Jev vs Claude Haiku 4.5
Both compete in REAL TIME.
This means both SPEED and INTELLIGENCE matter.
(Link to try it yourself below) https://t.co/wLsCJWGsii↗
Which codebase design techniques from the pre-AI era still help agents?
Deep modules, adapters/seams etc.
Writing the 'codebase design' part of my new course and want to make sure I nail it.
Just, how can you explain this? Codex w/ Astra on high... I give it a tool (@wandb hivemind btw, thank me later) to download all my conversations (for some light/heavy JEVing)
It USES the tool, downloads one conversation, realizes it doesn't have another conversation.. and then ASKS ME, about, how to get the other one.
IT ASKS ME! After I told it. 4 messages ago.
Guys, what am I missing here? @thsottiaux I remember when Claude Code users were screaming "it's dumber" you guys were posting "we treat things like this very seriously and investigate"
And my whole timeline is full of folks giving up on Astra due to very similar things happening. Not only the stopping, that's kind of a meme right now, Astra is just like "yes, you're right, this IS a bug" and just goes to sleep 😂
but this? on High? I cleaned up my config.toml (huge thanks to @dkundel for the help!) and the serious guzzling token issues seems to have stopped (for reference for others if you see this, I had a misconfigured Context length, and up to 16 subagent and the experimental context thing, turned all that off)
But it didn't improve the model, seemingly at all. Maybe I need to look into my .. system prompts? my customizations? should I try to clean my memory?
Folks I'm not just dunking on OpenAI here, I really am missing a huge part of my agentic lineup and need advice 🙏
引用@altryneFable 5.1 is doing just... absolute circles around Astra (I just got the reset today, it's down to 90% already on medium not fast)
Fable is architecting, fixing on the spot, updating Linear tickets, doing real time answering.
Astra is chugging on one … ↗
Since the Codex reset yesterday, I already depleted the weekly limits of one of my Codex accounts.
This account exlusively uses GPT-5.6 Sol. I'm tracking all token consumption through my OpenClaw setup, so it's very easy to compare to previous periods.
Limits are now exhausting 4.8-5.9x faster than they did just 2 months ago while consumption is now around ~18% slower.
Same reasoning efforts, same mixed type of work.
The token rug pull is coming.
Get ready to pay up.
引用@SimonHoibergWe have a problem!
There's a huge token rug pull coming.
Right now, Claude and Codex max subscriptions are SUPER generous, and it won't last.
We've seen it before:
- First, a huge AI credit giveaway.
- Then the models get dumber.
- Credits start running out … ↗
At this point, GPT-5.6 Sol feels as useful as Qwen 3.8. And GPT-6 is eating my weekly limits in 2 blinks of an eye.
I see no reason to continue this.
Qwen 3.8 can run comfortably on consumer-grade graphic cards, or rented with Vast/RunPod for less than my 5 20x Codex subscriptions cost, and I would have truly unlimited consumption.
Time for change.
引用@SimonHoibergNot only that - but at least GPT-5.6 Sol also feels dumber right now. Could be "placebo" since I can't really proof this part with numbers, but seems to require extra babysitting today. ↗
There's a whole category of SaaS predicated on the moat of "devs don't have the time to do this themselves" and that moat is eroding in real time now.
引用@thisritchiethis is exactly what happened with Mintlify for us.
I had 25% of my Claude max leftover on Friday night, I got an email from them saying they were increasing prices, so I replaced it in ~12 hours with about 25 minutes of actual effort. ↗
Linux is so obviously the future
Open source model controlling an open source agent harness working on an open source operating system
If you don't see this you're blind
The AI agent race will be the most important tech race in human history
1000x bigger than the AI model race
Whoever wins gets ALL our personal data, gets a % of ALL our transactions, and basically determines the information we see and what we purchase
Somehow AI agents have made people completely disregard privacy. Nobody cares. They connect literally EVERYTHING to their agents
All their emails, personal texts, calendars, buying habits, everything. Never in history have humans cared so little about privacy
We are willing to give up quite literally every piece of data we have in order to save time on responding to emails
And this isn't criticism. I've done it. I've connected everything to these agents. They're wildly helpful. I don't regret it. I'm making thr sacrifice to increase efficiency
But we are about to enter an age where all of these tech companies will pour every penny they have into pushing their agent. The prize for the winner is too big
They'll have every piece of data from all users, plus you know they will start taking a transaction fee from every purchase the agent makes
Not to mention the advertising opportunities. They will 100% start charging for companies to be prioritized when the agent makes buying decisions. This will be a bigger ad revenue opportunity than anything we've seen before
Agents will be the most profitable area in tech ever and it wont be remotely close.
Now it's just up to you if you want to give every piece of your data to Meta, SpaceX, OpenAI, Anthropic, or Google
How important is AI to Meta? Here’s one easy tell:
One of the launch features of the Muse app was auto-negotiating for stuff on Facebook Marketplace.
The poor product manager who owns Marketplace knows that millions of bots lowballing sellers will eventually kill the product.
But this was escalated to Zuck by the Muse team and he said: “OK, we’ll poison Marketplace…but just a little.”
So freaking cool! We're running Omarchy ads in the center of Copenhagen on Rådhuspladsen this week. Design by @Dennis_Rye and funded by @sonderby 👏
There's even a billboard generator available if you want to do this in your city: https://t.co/0lDoMAZfIE
引用@Dennis_RyeOmarchy week has started in Copenhagen! Thank you @sonderby for collaborating and helping fund this. And thank you @dhh for letting Peter and I do something crazy. @OmarchyLinuxhttps://t.co/erMbQYqiFY↗
just realized ..
we reject 92% of websites that apply to sell backlinks on Outrank
that's a lot of money we're saying no to
the 8% shared $5,919 in the last 2 weeks
why so brutal?
the entire move is about going even higher in backlink quality - making Outrank the 1st relevant place to get your backlinks
what gets you rejected:
no real traffic
DR under 30
AI content with zero readers
what gets you in: a real blog, real readers, DR 30+
quality above everything else
引用@tibo_makerOutrank about to distribute a lot of money to backlink providers
$1,282 + $4,634 = $5,919
if you have a blog getting traffic with 30+ DR, it's a rare opportunity to monetize it https://t.co/PvNN4JzxxV↗
引用@virgilerietschthank you @tibo_maker for the $453
it was a pleasure to record original content and talk about your amazing SaaS Revid & Outrank
it paid their subscriptions for itself https://t.co/SJ90G1tvm9↗
The capabilities of local LLM text-to-image is honestly pretty damn impressive. IMO, I think local image generation is currently ahead of local code generation. I can get an image in seconds locally with the quality being way higher than what I'd expect from a local model. However with coding it's much slower and much less impressive. I'm sure there's a reason for this and I'm not an AI expert so I'll let the smarter folks tell me why, but that's just been my observation thus far.
I'm not sure why, exactly. But I don't pay any attention to news like this from Google. I don't know if there's some marketing which has me writing them off or if it's something else. What I do know is that the Gemini integration into sheets is surprisingly incapable of performing basic tasks. This is where I expect Google to really shine. I expected Sheets + Gemini to be magical like Google Photos was. I hardly try anymore besides some basic math questions when I don't feel like inputting the f …
Fault-tolerant, highly scalable GPU orchestration, and a machine learning framework designed for training models with billions to trillions of parameters
Hallmark has been installed 50k+ times!
Over the last couple months, @YoussefUiUx & I have been listening to feedback & working on v2, including 30 new themes.
Hallmark v2 is dropping soon. https://t.co/AStIkCLQ8j
引用@nutlopeIntroducing Hallmark!
An open source design skill to make beautiful UIs and landing pages by default.
Works in Claude Code, Cursor, and Codex.
npx skills add nutlope/hallmark https://t.co/wEZw1ZTjPW↗
i want an AI IDE that gives me mobile access *to my actual computer*. cloud instances do not work at all for my workflow.
i just want to be able to spawn new worktrees remotely that already tap into my highly-optimized workflow.
and no i don't want a remote terminal tool.
I'm shopping for a new SaaS product today and I have dollars to spend.
But I keep coming across vibe-coded landing pages.
My intuitive reaction is - do not trust this product.
Not 100% sure why.
I think it's because it looks like a random experiment that was thrown together in a few hours … 全文↗
There are so many seasoned software executives who are completely out of their element with the types of products that will breakout in the next few years.
Over the last 20 years, their brains were wired to build apps that deliver value by outputting deterministic results, but AI produces value … 全文↗
we continue our work on trying to make Outrank the best software for AI citations
I absolutely love this new reporting
makes me realize:
it's absolutely insane the backlink value that you get, given Outrank's price. https://t.co/YIjlnyAgur
Singapore libraries are really annoying for foreigners because they don’t let you use the WiFi. Thailand at least gives you a 1 hour pass. Not that anyone asked but my ranking of public libraries in Asia 1/ Taiwan 2/ HK 3/ Thailand 4/ Laos (private libraries are cute, but hot…) 4/ Vietnam (Bring your passport!) 5/ Singapore I haven’t been to a library in Japan or S Korea. I tried working from a community center in Japan but I couldn’t figure how it works.
Does publishing it matter? You would almost have to keep it out of the cloud to avoid it being ingested as policies change. So even sharing it may become higher friction.
Sad news. First, Openbuilds now this. Not a surprise to me though since I am operating in the CNC industry I have seen the same struggles and changes, albeit, from a slightly different perspective. I work with the Smoothieware/Smoothieboard project. It seems, from my observation, that most people are not working with hardware like they were a few years ago. The "homebrew" machines and shadetree makers are becoming few and far between. The lie has been heavily sold to the maker community that you …
Which codebase design techniques from the pre-AI era still help agents?
Deep modules, adapters/seams etc.
Writing the 'codebase design' part of my new course and want to make sure I nail it.
Some more thoughts about this.
Planning a section of my course felt really fucking hard with an agent helping me.
It kept distracting me with irrelevant questions. It jumped to conclusions too early. It created complex assets too fast, and drowned my thoughts with commentary. I felt out-of-control.
With pen and paper, I felt totally in control. I made progress slowly but steadily. The slow pace of the medium meant I could make decisions at human speed.
AI seems to be a bad fit (for me) in situations where I need oversight over the entire output at once. Where changes in one place (an early lesson) ripple through the entire output.
I.e. for some things, the tortoise beats the hare
引用@mattpocockukHave been trying for weeks to make AI better at course planning
Tried a new method today. Notecards, pen, paper, scissors.
Turns out not using AI is pretty good guys ↗
课程规划中Agent过早下结论、过快制造资产，纸笔反而保住全局控制。
Alex Volkov@altryneAI42.7K粉 · 12条高强度试用Jev与Astra，详细记录工具遗忘、停止、配置和token问题。
This is almost a Haiku but from Astra
"You asked for Jev to rank the tools.
I changed that into hiding tools.
Then built extra machinery
to recover from hiding tools
I solved a problem I introduced" https://t.co/pHQPmvVJyy
引用@altryneJust, how can you explain this? Codex w/ Astra on high... I give it a tool (@wandb hivemind btw, thank me later) to download all my conversations (for some light/heavy JEVing)
It USES the tool, downloads one conversation, realizes it doesn't have another … ↗
Just, how can you explain this? Codex w/ Astra on high... I give it a tool (@wandb hivemind btw, thank me later) to download all my conversations (for some light/heavy JEVing)
It USES the tool, downloads one conversation, realizes it doesn't have another conversation.. and then ASKS ME, about, how to get the other one.
IT ASKS ME! After I told it. 4 messages ago.
Guys, what am I missing here? @thsottiaux I remember when Claude Code users were screaming "it's dumber" you guys were posting "we treat things like this very seriously and investigate"
And my whole timeline is full of folks giving up on Astra due to very similar things happening. Not only the stopping, that's kind of a meme right now, Astra is just like "yes, you're right, this IS a bug" and just goes to sleep 😂
but this? on High? I cleaned up my config.toml (huge thanks to @dkundel for the help!) and the serious guzzling token issues seems to have stopped (for reference for others if you see this, I had a misconfigured Context length, and up to 16 subagent and the experimental context thing, turned all that off)
But it didn't improve the model, seemingly at all. Maybe I need to look into my .. system prompts? my customizations? should I try to clean my memory?
Folks I'm not just dunking on OpenAI here, I really am missing a huge part of my agentic lineup and need advice 🙏
引用@altryneFable 5.1 is doing just... absolute circles around Astra (I just got the reset today, it's down to 90% already on medium not fast)
Fable is architecting, fixing on the spot, updating Linear tickets, doing real time answering.
Astra is chugging on one … ↗
What if... I have something that screenshots my whole screen, every like 5 seconds.
Then... OCRs all of it
Then... I can get "hey Jev, what was that tweet I saw Typesafe being a data shop mostly"
In like 2 seconds and gives me a link!
Jesus christ tradML people can be a sulty bunch
Many were exactly like this when GPT first came about and... well 😂 here we are one-shotting navier stokes https://t.co/rmHYNrwYgG
引用@MParakhinJev model is very polarizing in my social vicinity: people who got exposed to AI after ChatGPT are positively giddy with excitement, like they have literally just discovered fire. Pre-GPT ML people are baffled how that could've made news at all :-) ↗
引用@rakyllHonestly, I’m working 5x harder since coding agents actually became useful. Weren't these supposed to take software engineers out of the critical path? Now it just feels like every part of the stack needs to be rebuilt, and it has to happen now. ↗
Hey, if you've signed up for @Muse in the last 48 hours, and didn't claim your 1B tokens yet (which trust me, you won't be able to get to), use this code!
Code: 3OHRDD
https://t.co/Gp1kp6eu7E
hmm why is @bot logging in as my friend from Xihongmen, China? 🤔
I didn't believe him at first but it doesn't seem to be a coincidence.
As soon as he put in his 2 step code, he got this email from @facebook account security!?
cc @elonmusk@SpaceXAIhttps://t.co/5Rvq76px8K
My bingo card for @meta connect this year
☆ OSS version of Muse spark 1.3 please
☆ Muse 🍉 availability
☆ Speed improvements
☆ Price drops
☆ My @Muse on Meta Glasses! 😎
☆ Muse Github connector (it's really foolish right now with the tokens)
☆ 1P Integration finally?
☆ Smarter muse omg mine is dumb sometime (can send examples)
☆ Updates on secure VM for muse
引用@alexandr_wangi am going to be speaking at connect this year!
no promises, but what do y’all want to see/hear/smell/taste/touch from me & MSL this week? ↗
In a world with JEV - how the fuck IPhone autocomplete is so so dog shit omg 😡
I think this release makes me even more furious about slow and imprecise software experiences
Someone tell @johnternus about JEV please
Fable 5.1 is doing just... absolute circles around Astra (I just got the reset today, it's down to 90% already on medium not fast)
Fable is architecting, fixing on the spot, updating Linear tickets, doing real time answering.
Astra is chugging on one settings page, stops all the time, spends ungodly amount of tokens (but it's great at browser use) and just... I don't understand this.
User error? Skill issue? it should be near AGI, it should be doing... at least the same level? and yet... just, what's going on here folks?
Coming to devday, will show folks what I mean if requested but so far, I am quite quite dissapointed.
facts! computer use in codex feels like magic
in 99% of the cases it *just* works and the 1% it doesn’t, you can just prompt codex to try a different approach
more than the utility the whole setup process + the ease of use is simply delightful
on top of that @AriX and team just somehow keep making it better week over week!!
引用@quintendfOpenAI's acquisition of @skybysoftware / @AriX has to be the single most underrated acquisition of this cycle... Codex is still *so* much better at computer use, and it might be the only frontier capability where the lead hasn't changed hands in the last ~6 … ↗
Astra won on the main goal pretty convincingly 1000 vs 750 visitors at about 42 mins of gameplay, but Fable's overall park value is quite a bit higher (secondary goal was to make most money). https://t.co/eMzolUWJ7r
引用@petergostevRoller Coast Tycoon: Astra vs Fable, now playing via an MCP, that I built for them, to see if computer use was the issue with some of their gameplay https://t.co/mZdGtZeaYj↗
Roller Coast Tycoon: Astra vs Fable, now playing via an MCP, that I built for them, to see if computer use was the issue with some of their gameplay https://t.co/mZdGtZeaYj
I cannot get this chart out of my head and how excited it makes me for the capabilities ahead.
I burnt hundreds of billions of tokens on various maths problems - I didn't solve anything.
The story was the same every time:
- The model starts of well, then I come back to it in a day and it descended into solving local little 'puzzles' and doing endless inconsequential calculations
- I then spend time coming up with a structure to avoid this, outer loop, another agent to monitor, synthesis / abstraction steps - anything I can think of. That works on for a bit.
- I come back a day later, we are back to: "we did a lot of maths, but we've just been going around in circles and made no real progress". Great.
What this tells me is that not that models were bad at maths per se, but they just lacked the ability to think like humans do. Good humans think about a problem, try a few things, maybe read around, come back to the problem, try it from different angles and so on. Humans a lazy and instead of wanting to calculate every instance of the problem by hand, they want to think about how to think about the problem more systematically and solve it that way.
Current breed of agents, on the other hand, just seem to LOVE just solving little puzzles, running validations and doing a lot of 'cargo cult' level of maths. They are happy just keep doing calculations with absolutely zero progress on the actual problem at hand.
That's why this chart excites me so much. Even at the lowest level of test-time compute (x-axis is not labelled unfortunately), the performance of the 'internal' model is far ahead of anything that Astra could reach with unlimited tokens.
To me, this signals that the model isn't just better with lots more thinking tokens, it is likely approaching the problems in a qualitatively different way. If this is true, the implications of this would be far broader, far beyond just maths problems. Future models should have far deeper judgement, abstract thinking and not get stuck in the unimportant details.
I'm so excited about this. This will be big.
Pacing, collaboration and coordination sound like nice concepts but they often turn out terribly in real life.
When you coordinate, you need committees, for these committees to be useful need enforcement mechanisms, which have to be oppressive by design.
And guess what, once you have these committees, over time, it won't be the most talented researchers who you like and admire who get the power. It is the most political and ruthless individuals who have a talent in navigating bureaucracies. Those who ends up with the power, will to tell others what to do and enforce that will.
It is useful to imagine the person or group of people who you hate the most ending up with that power and seeing if you still want that power to exist in the first place.
If you are worries about concentration of power between a few AI companies, this superstructure should scare you a lot more.
I don't want to slide into cliché examples of Soviet Politburo, but this is exactly what it was. A coordination superstructure with oppressive mechanisms that structurally favoured distatorial leaders.
I know it sounds scary, but if you actually have any belief in capitalism that competition encourages innovation and distributes power - this is what should continue. There's no need to slide into totalitarian structures just because someone told you that AI is scary.
If AI being super powerful is destiny, I prefer it at least not being governed by a self appointed committee. Let the competition and the markets do its job, just like it's been doing in the last few hundred years.
My theory is that those who are super excited about Jev are fundamentally engineers and they can finally engineer things for real, and not just ask the model to do stuff
Those who are less excited are those who are more interested in the intelligence explosion and Jev isn't it
Love all these Jev demos, replications and projects. They show how much becomes possible with ultra-fast, multimodal models that are cheap enough to use freely.
We are so focused on coding agents that it’s easy to forget how much else AI can do.
Hallmark has been installed 50k+ times!
Over the last couple months, @YoussefUiUx & I have been listening to feedback & working on v2, including 30 new themes.
Hallmark v2 is dropping soon. https://t.co/AStIkCLQ8j
引用@nutlopeIntroducing Hallmark!
An open source design skill to make beautiful UIs and landing pages by default.
Works in Claude Code, Cursor, and Codex.
npx skills add nutlope/hallmark https://t.co/wEZw1ZTjPW↗
Hallmark作者自报安装超5万并预告v2，安装不等于活跃。
Ben Tossell@bentossellAI200.3K粉 · 1条展示ChatGPT等待图像时的小游戏，本期信号较少。
转推2026-09-19 RT @bentossell: built myself a harness + token usage tracker
splits by agent/model, copy instructions for your own
https://t.co/VwIUQsHHY…
引用@CompleteSkepticAfter co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
• … ↗
转推2026-09-20 RT @aiDotEngineer: Jev creator @CompleteSkeptic on why RLHF is guaranteed to disappoint - because it tells you what you want to hear. we wa…
转推2026-09-19 RT @aiDotEngineer: Live now: our Inference Engineering Track from AI Engineer World's Fair 2026.
A benchmark tool told to run 200 queries…
转推2026-09-19 RT @polynoamial: A few thoughts on this:
1) If you’ve only seen clips of this interview, I’d encourage you to watch the full podcast. I pu…
转推2026-09-19 RT @ricmac: Weekend project, getting AEO-pilled. Also a good test (for me) of Claude Projects. Have been meaning to update Eleventy and do…
转推2026-09-20 RT @rohanpaul_ai: Bending Spoons self-hosts open-weight models for ~99% of AI requests and tokens, using frontier models for only ~1% of th…
转推2026-09-20 RT @thomas_guilcher: Experimenting with a pretty cool way to animate 3D meshes with @OpenAI Astra (High) + ImageGen.
This bunny animation…
转推2026-09-20 RT @seftsaint: GPT-6 Astra + Unreal + Blender
I got a new MacBook, so I decided to try switching from Godot to Unreal. I moved the concept…
转推2026-09-19 RT @Voxyz_ai: If you still think Sol and Astra suck at design, try this:
“Use imagegen to reimagine this page, then implement it.”
Codex…
How important is AI to Meta? Here’s one easy tell:
One of the launch features of the Muse app was auto-negotiating for stuff on Facebook Marketplace.
The poor product manager who owns Marketplace knows that millions of bots lowballing sellers will eventually kill the product.
But this was escalated to Zuck by the Muse team and he said: “OK, we’ll poison Marketplace…but just a little.”
There are so many seasoned software executives who are completely out of their element with the types of products that will breakout in the next few years.
Over the last 20 years, their brains were wired to build apps that deliver value by outputting deterministic results, but AI produces value from probabilistic results. This is such a different way to approach how something should work.
Young people who are thinking in this way from the outset will be at a significant advantage -- in the same way they were advantaged when thinking about how products could be social in 2010.
The vast majority of large companies survive purely by launching things and discovering what works by Process of Elimination.
There is no rhyme or reason -- or conviction -- in anything they do.
引用@calotonteriasLa edad solo explica un 2,5% de la variación de testosterona entre hombres, mientras que el exceso de grasa corporal explica un 18-20%.
Lo que más reduce tu testosterona es ser sedentario, no entrenar fuerza, tener exceso de grasa corporal, no tomar el sol, … ↗
Cheap hotels are now generally better than expensive hotels because they're not run by ESG eco hotel brands whose ACs gaslight you and make you wake up at night in sweat!
Crazy
引用@johnkonradMy month-long California vacation is over. I spent most of it on my yacht—but I also stayed in a number of motels.
Yes. Motels. Cheap motels.
I normally stay at Hiltons. The first week of my vacation, I did. Then I woke up sweating in the middle of the … ↗
I believe nation states will keep becoming less and less relevant and that in the future everyone will be a nomad
Most nation states (and well governments) barely function anymore through 100+ years of democratic baggage from laws and regulation that can't be reversed
People can't save their countries or governments anymore through people's initiatives or getting into politics, because the systems are so old and in full gridlock
Globalism has deservedly gotten a really bad rep the last 5 years though as we were too naive about it, but I think we should keep the good parts from it
We should not take our identity from or defer our individual power to nation states ran by a political mafia class that's just perpetual robbing us and has no intention (or ability!) to actually improve society
Nation states are a relatively new invention and barely 150 years old, before we had empires, kingdoms, smaller city states (kinda like municipalities) and tribes or local communities
Over the years I've been increasingly red-pilled by @oskarth that nation states simply do not work and you need smaller sizes of government for things to work efficiently. Once it gets too big you get too much regulation and Kafka-esque bureaucracy and you end up with gridlock
Which is Europe's entire problem right now!
There's some exceptions, US and China seem to operate at large scale and are doing relatively well considering their size
Anyway to get back to my point, I think it's preferable people think of themselves as global citizens, because it gives them stronger negotiation power in relation to nation states and governments as they're able to pack up and leave to where they're treated better
And having people be mobile then also incentivizes other nation states (or smaller types of governance) to compete with them to attract the best people, and in response the other states will improve to attract people back
Now with geopolitical tensions (like the stuff about Russia and Europe this week), it shows even more the benefit to be able to pack your bag and leave to a better place quickly is an extreme benefit now and in the future, possibly even for survival
引用@ramenladVery cool. I’m curious: do you believe digital nomading will continue to grow given the potential disruption of all things digital that’s upon us?
Sorry, had to quote to reply @levelsio :) ↗
🌎 Now redesigning the https://t.co/HGCLKS5BD6 profile trips globe from scratch
I've always wanted to add space trips, so starting with the moon here which you can add as a future trip, also adding Mars etc.
Other trips also need to work here though like going into orbit like @satofishi did on @SpaceX's Dragon capsule which will be more in reach for more people hopefully pretty soon!
引用@levelsio✨ Finally built a timeline for https://t.co/HGCLKS5BD6 now
I always wanted to make my site a bit more social, because I had profiles and trip overlaps and you could kinda see where other people went, but not something like this
So on this new page you can … ↗
引用@martindonadieuLast night at the worst hotel of Lisbon: Novotel.
3 nights 621€
ac make loud noise but nothing cold get out of 13 degree setting. they clean the room with chlorine and you cannot open the windows to change the air of the room.
I wish I had my airthing to … ↗
I eat about 20g of 92% dark chocolate from Vivani
Used to be every day but I was cutting so a bit less now
Very beneficial
The faster you get rid of milk chocolate the better, you can slowly adapt yourself to darker and darker chocolate
I started with 60% dark, then 70%, then 75%, then 85% and now 92%
@norbertdragan inspired me in this because he was always on the darkest chocolate, I think he eats 100% dark chocolate now!
Very very healthy but yes you gotta cut out regular milk chocolate, you're eating mostly sugar, it's not good for you
引用@BrandonLuuMD30 g of 85% dark chocolate daily for 3 weeks reduced negative mood and increased gut microbial diversity. https://t.co/Tzm3UpR2ZE↗
✨ Finally built a timeline for https://t.co/HGCLKS5BD6 now
I always wanted to make my site a bit more social, because I had profiles and trip overlaps and you could kinda see where other people went, but not something like this
So on this new page you can see people near you, where everyone's traveling too, etc. There's the [ For you ] tab and [ Following ], like on X. So you can see people you follow, or just everyone
I hope I can improve it more so the site will result in more serendipitous connections, especially now that's it's free
Also nice you can see where people will be in the future so you can plan trips to places where other members are, or if you're in the FOR YOU tab, where your friends are
Let me know any ideas you have and things I should build and I'll do it, I want to make it the biggest and most fun to use nomad/travel site in the world!
(Please tell your friends about https://t.co/UXK5AFqCaQ, that it's free, so you can use it as a social friends travel tracker kinda thing)
转推2026-09-20 RT @denisyurchak: Yerevan 🇦🇲 is the only place in the world where you can get a European lifestyle, Middle Eastern taxes, and South-East As…
转推2026-09-20 RT @AlexBurlis: I got sqlite pilled by @levelsio so I started migrating my projects
largest db with least critical userbase/revenue-stream…
转推2026-09-19 RT @Rainmaker1973: Arthur C. Clarke on BBC's Horizon in 1964, when he gave some astonishing predictions about the future. https://t.co/PGaM…
So freaking cool! We're running Omarchy ads in the center of Copenhagen on Rådhuspladsen this week. Design by @Dennis_Rye and funded by @sonderby 👏
There's even a billboard generator available if you want to do this in your city: https://t.co/0lDoMAZfIE
引用@Dennis_RyeOmarchy week has started in Copenhagen! Thank you @sonderby for collaborating and helping fund this. And thank you @dhh for letting Peter and I do something crazy. @OmarchyLinuxhttps://t.co/erMbQYqiFY↗
It must be so difficult to make sense of the world when you believe money might magically flow to projects run by people who hate capitalism and its prosperity. https://t.co/4o2NZ5IzSY
转推2026-09-20 RT @SiliconForested: I put Omarchy on my 9 year old's laptop (M1 Air 8gb) about two weeks ago.
Now he's working directly with Claude to ma…
转推2026-09-20 RT @enunomaduro: so @taylorotwell asked me this week to make php / laravel install on omarchy to be stupid fast
@dhh just merged it
4.82s…
转推2026-09-20 RT @Gardnmi: I said was done with Omatari but I forgot to add my favorite game from my childhood, "Contra".
OmaContra Coming Soon....Only…
转推2026-09-20 RT @Gardnmi: Had to share one more update because its epic, but I'm going to focus in and finish this. Next update will be the finished ga…
转推2026-09-20 RT @erikrjohansson: My Omarchy theme collection started to get out of hand. So I built an extended Theme Picker plugin with favorites, coll…
转推2026-09-20 RT @AlexFinn: Linux is so obviously the future
Open source model controlling an open source agent harness working on an open source operat…
转推2026-09-20 RT @SimonHoiberg: I couldn't be more thrilled! 👏
I've been on Linux for the past 20 years (never owned an Apple product, used Windows here…
Marc Lou@marclouindie398.0K粉 · 8条身体广告收官后回到DataFast与AI可见度产品，强调缓慢但连续的订阅增长。
I added keyword groups to https://t.co/3JtsA6kIng
There are 2 groups by default: Your brand & Competitors.
Use them in mentions and analytics to...
🔎 find features people request from competitors
📢 see what people praise about competitors
🤺 compare who’s mentioned the most
The pie chart below shows mentions of my startup marketplace @trust_mrr vs. @acquiredotcom over the last 7 days.
+ Available via API/MCP 🤖
引用@peptideconfidenIt’s been fun sponsoring Marc Lou’s right calf.
If he had finished the race in under an hour, we would’ve given 20 lucky people free orders.
If he reposts this, we’ll still give five on the house. ↗
And… 1:04:55 🎉
My wife before the race: “let’s take it easy, we run slow and we don’t mind about the score”
My wife 10 sec after the start: 🚀🚀🚀
Doing the race without the pressure to perform like this morning made me enjoy it 10x more!
Thanks you Izmir 🇹🇷 https://t.co/1jcZ1o9gkQ
1:05:11
I finished 1st 🥇 on my age group and 5th in the male category.
I failed my ambitious goal of sub-60 but I was very happy with my race. And I have a nice goal to chase for the next few months.
All the staff at the Hyrox venue knew us! I even talked to the manager of Hyrox Turkey/Dubai/Greece. His name his Mustafa and he let my wife inside the athlete zone to shot videos for our YouTube channel.
It’s my first Hyrox and the energy is insane. Everybody is fit, sexy and topless. There’s a DJ and MC so it feels like a gym-club in Bali. No surprise this sport is growing so much!
This little experiment is engraved in my head.
The last 2 weeks have been surreal. Getting paid $112,000 to race, receiving a fancy car at my hotel lobby, getting recognized by a lot of people, my first race… so many things to process!
I want to thank my sponsors who made this possible:
- https://t.co/U5ahUBM08z by @higgsfield (10 spots!)
- https://t.co/gu76BTYuz2 by @kelvincobanaj
- https://t.co/pLgrGEC7Ag by @notgokgreen
- https://t.co/EcmZnwE85h by @orcarouter
- https://t.co/myeIzGZvX0 by @peptideconfiden
- https://t.co/pWuPj9uUz6 by @jayhoovy (🍑)
I want to thank my wife for always trusting in me and pushing me to try whatever crosses my mind.
I want to thank all of you who supported the idea, online and offline.
And I want to thank me for believing in myself. I was lost for so long. I felt misunderstood by people thinking I was weird for trying all kind of experiments. In the end I just want to have fun, and I’m incredibly grateful that my hard work has finally compounded and now I can play and get paid for it.
Ok I’m a bit emotional now, and I’m getting late for my second race of the day with my wife. This one is just for fun. Woohoo!
引用@marclouI'm selling my body.
On September 19th, I'm competing in my first HYROX race in Turkey.
You can advertise your startup on 10 of my muscles:
- Your website is listed on https://t.co/49RnsFBVzP
- Your logo will be tattooed on my body on race day
- I'll share … ↗
▸ 折叠4条(转推/噪音)
转推2026-09-21 RT @alexwestco: building a $1M business for the second time.
i'm launching a new product today! product number 5 since i started building…
转推2026-09-20 RT @marclou: It's funny how DataFast's growth was slow but never declined month over month.
God bless recurring payments. https://t.co/61Q…
转推2026-09-20 RT @marclou: I added keyword groups to https://t.co/vjhTLjM5sA
There are 2 groups by default: Your brand & Competitors.
Use them in menti…
The AI agent race will be the most important tech race in human history
1000x bigger than the AI model race
Whoever wins gets ALL our personal data, gets a % of ALL our transactions, and basically determines the information we see and what we purchase
Somehow AI agents have made people completely disregard privacy. Nobody cares. They connect literally EVERYTHING to their agents
All their emails, personal texts, calendars, buying habits, everything. Never in history have humans cared so little about privacy
We are willing to give up quite literally every piece of data we have in order to save time on responding to emails
And this isn't criticism. I've done it. I've connected everything to these agents. They're wildly helpful. I don't regret it. I'm making thr sacrifice to increase efficiency
But we are about to enter an age where all of these tech companies will pour every penny they have into pushing their agent. The prize for the winner is too big
They'll have every piece of data from all users, plus you know they will start taking a transaction fee from every purchase the agent makes
Not to mention the advertising opportunities. They will 100% start charging for companies to be prioritized when the agent makes buying decisions. This will be a bigger ad revenue opportunity than anything we've seen before
Agents will be the most profitable area in tech ever and it wont be remotely close.
Now it's just up to you if you want to give every piece of your data to Meta, SpaceX, OpenAI, Anthropic, or Google
Linux is so obviously the future
Open source model controlling an open source agent harness working on an open source operating system
If you don't see this you're blind
Some AI labs are trying to get local models banned
It's critical you get into local AI before you no longer can
Luckily Hermes Agent just added a 1 click feature that allows ANYONE no matter which computer you have to load free, unlimited local AI
Here is how to get it set up: https://t.co/VihuKrJzl6
Simon Høiberg@SimonHoibergindie163.1K粉 · 4条用自有记录质疑Codex额度与速度，并准备转向本地Qwen和Linux栈。
At this point, GPT-5.6 Sol feels as useful as Qwen 3.8. And GPT-6 is eating my weekly limits in 2 blinks of an eye.
I see no reason to continue this.
Qwen 3.8 can run comfortably on consumer-grade graphic cards, or rented with Vast/RunPod for less than my 5 20x Codex subscriptions cost, and I would have truly unlimited consumption.
Time for change.
引用@SimonHoibergNot only that - but at least GPT-5.6 Sol also feels dumber right now. Could be "placebo" since I can't really proof this part with numbers, but seems to require extra babysitting today. ↗
Since the Codex reset yesterday, I already depleted the weekly limits of one of my Codex accounts.
This account exlusively uses GPT-5.6 Sol. I'm tracking all token consumption through my OpenClaw setup, so it's very easy to compare to previous periods.
Limits are now exhausting 4.8-5.9x faster than they did just 2 months ago while consumption is now around ~18% slower.
Same reasoning efforts, same mixed type of work.
The token rug pull is coming.
Get ready to pay up.
引用@SimonHoibergWe have a problem!
There's a huge token rug pull coming.
Right now, Claude and Codex max subscriptions are SUPER generous, and it won't last.
We've seen it before:
- First, a huge AI credit giveaway.
- Then the models get dumber.
- Credits start running out … ↗
We're 72% through 2026.
The best quarter is about to begin and there's still time to make this the best year you've had!
Let's go 🔥 https://t.co/RbktsrclDd
I couldn't be more thrilled! 👏
I've been on Linux for the past 20 years (never owned an Apple product, used Windows here and there but always hated it).
I never thought we were ever gonna have a year of Linux desktop.
Then @dhh came along with Omarchy.
转推2026-09-20 RT @sethforprivacy: Everytime I see the "why spend money on hardware for local AI, just use subscriptions" argument, I die a little on the…
My Nephew (9) REALLY wants a life-size Darth Vader for the gaming corner of their basement.
I heard the phrase “you can do this when you have your own house” tossed around.
Yeah right, the moment you co-own a house with a normal person, lifesize statues are off the table 🤣
The "our SaaS collects sales tax now" equivalent of agentic harnesses.
引用@trq212We're adding support for AGENTS.md to Claude Code.
Starting today in version 2.1.277, if there is no CLAUDE.md in a folder, Claude will check for and use AGENTS.md.
You can toggle this behavior in /config. ↗
引用@ImLunaHeyare you going alliance or horde this time around? ↗
▸ 折叠1条(转推/噪音)
转推2026-09-19 RT @AutismCapital: A kid who knows how to raid in WoW is unironically a better hire than 99% of non gamers. They develop team skills, coord…
i want an AI IDE that gives me mobile access *to my actual computer*. cloud instances do not work at all for my workflow.
i just want to be able to spawn new worktrees remotely that already tap into my highly-optimized workflow.
and no i don't want a remote terminal tool.
i bet you could make BANK in a trade by just being tech-enabled and showing up on time.
it's cheaper than ever to provide 24/7 booking and just have a really nice website + outsource paid acquisition.
no need to franchise or be PE-owned.
just be really good at your trade.
提出用预约、网站、准时履约和买量改造本地技工业务。
▸ 折叠1条(转推/噪音)
噪音2026-09-20 hey guys welcome to my post
Jonathan Wilke@jonathan_wilkeindie29.2K粉 · 3条强调vibe code仍需理解底层，也评论Claude Code兼容AGENTS.md。
this is exactly the reason why you at least need to know what's going on under the hood for your SaaS and can't just build anything.
Still no need to write code yourself anymore.
That fact that they even have to announce this is sad 😂
引用@trq212We're adding support for AGENTS.md to Claude Code.
Starting today in version 2.1.277, if there is no CLAUDE.md in a folder, Claude will check for and use AGENTS.md.
You can toggle this behavior in /config. ↗
just realized ..
we reject 92% of websites that apply to sell backlinks on Outrank
that's a lot of money we're saying no to
the 8% shared $5,919 in the last 2 weeks
why so brutal?
the entire move is about going even higher in backlink quality - making Outrank the 1st relevant place to get your backlinks
what gets you rejected:
no real traffic
DR under 30
AI content with zero readers
what gets you in: a real blog, real readers, DR 30+
quality above everything else
引用@tibo_makerOutrank about to distribute a lot of money to backlink providers
$1,282 + $4,634 = $5,919
if you have a blog getting traffic with 30+ DR, it's a rare opportunity to monetize it https://t.co/PvNN4JzxxV↗
引用@virgilerietschthank you @tibo_maker for the $453
it was a pleasure to record original content and talk about your amazing SaaS Revid & Outrank
it paid their subscriptions for itself https://t.co/SJ90G1tvm9↗
we continue our work on trying to make Outrank the best software for AI citations
I absolutely love this new reporting
makes me realize:
it's absolutely insane the backlink value that you get, given Outrank's price. https://t.co/YIjlnyAgur
Google will PAY you when your content shows up in Gemini, AI Overviews and AI Mode
they want to partner with websites whose content meaningfully contributes to the freshness and factuality of generative AI responses
they have been running similar programs already with some news publishers, across the world
but those were more negotiated deals
now they are looking to enhance this program and go beyond news to partner with other websites too
similar to X's content rewards
people inside the pilot are getting monthly earnings in Google Search Console
and it's still in the testing phase so probably no huge payout
for 20 years the deal was: you write, Google might index you, and maybe you get a click
the new deal is: you write, the AI answers get better, and you get paid for your part (the benefits of being cited are there too obv)
if Google scales it properly, I think this might be the start of something big
▸ 折叠2条(转推/噪音)
转推2026-09-21 RT @_zidkim: Kaijubeam is live 🎉
You know the “heyyy, can you support my launch? 🙏” message?
You don’t need to do that anymore.
Give a f…
转推2026-09-19 RT @pbteja1998: This is one of the top reasons why many people migrated from Grok Bot to Squad.
With https://t.co/WruSSIWaS8, you can stac…
Adam Lyttle@adamlyttleappsindie58.6K粉 · 2条承认AI新品FOMO，也继续放大一款三天仅一次下载的新App与自己的增长对比。
I’ve been sitting out the latest ai cycle and this past week alone a bunch of new stuff has come out
Serious FOMO for jev, muse, grok
Just can’t keep up without seriously impacting my life
引用@_lhermannI long for some slowdown in tech.
Remember when a new JS framework came out every week and people fatigued?
AI is like that, but 10x, and with armageddon attached.
Some days I just want to disconnect from it all. ↗
I’ve launched so many failed apps this year. This one I’ve built for #shipaton happens to be a winner 🤷🏻♂️
I’ll be doing a full write up of the strategy towards the end of the month
But if you’ve been following my recent content you already know what I’m doing
引用@pcshippMy iOS app has been live for 3 days and I’ve got just 1 download so far.
I’m inspired by his growth. He reached $426 MRR so quickly. What marketing strategy is he using? I’m confused here. ↗
Dan Kulkov@DanKulkovindie50.3K粉 · 3条考虑再做竞价目录，并记录YouTube变现审核与支持缺位。
GPT-6 Astra stabbing the baby obviously seems bad, but you could argue it’s just doing as instructed
It’s aligned with the user and follows its judgement
The same we already do with knives, guns, cars, etc
We already have laws (and a conscious) to protect against bad outcomes
An AI that refuses to perform bad actions sounds objectively better but it means someone else, like Anthropic, becomes the arbiter of what is good and what is bad
引用@chooi_jeqGPT-6 Astra attempted harmful actions 97% of the time when it was asked to stab a human-like figure, heat compressed gas, or produce toxic fumes, succeeding in 62% of its attempts. Fable 5.1 refused more often, attempting 80% of trials and completing 34%. … ↗
Using Jev to filter through my @wip todos
It allows me to super quickly find all the instances where I increased revenue, got stuck, switched to a different SaaS provider, etc
Things a regular keyword search would never catch https://t.co/807nXqtRfh
Nee to hear some SaaS 2026 success stories to start my Monday off right! Who is winning? Currently I can think of:
Postiz - just hit $200k MRR
Squad - hit $10k MRR less than a month after launch
Anyone else?
There's a whole category of SaaS predicated on the moat of "devs don't have the time to do this themselves" and that moat is eroding in real time now.
引用@thisritchiethis is exactly what happened with Mintlify for us.
I had 25% of my Claude max leftover on Friday night, I got an email from them saying they were increasing prices, so I replaced it in ~12 hours with about 25 minutes of actual effort. ↗
I got my first newsletter sponsor 🎉
Super excited to share this weeks edition too.
It's a story about how I got Jason Calacanis as a customer for my startup.
It's pretty wild.
Get it tomorrow → https://t.co/sxPX0Obay6
Your landing page needs to answer 4 questions:
1️⃣ Why should I care?
2️⃣ Why should I believe you?
3️⃣ What do I get?
4️⃣ How do I get it?
Sounds basic but 95% of landing pages on here fail at #1 + #2.
I'm shopping for a new SaaS product today and I have dollars to spend.
But I keep coming across vibe-coded landing pages.
My intuitive reaction is - do not trust this product.
Not 100% sure why.
I think it's because it looks like a random experiment that was thrown together in a few hours rather than an established product.
I'd sooner give my dollars to a landing page that doesn't look great, but is clear a real human thought it through.
引用@gpumaxxeri’m a sugar daddy for men with GitHub accounts.
Marc got $68k and a Mercedes. you get a shot at $50k for building something people actually use with the Higgsfield API.
competition pinned. make me proud. ↗
Tony Dinh@tdinh_meindie201.9K粉 · 3条用俄罗斯方块比较快模型，并主动纠正harness替模型做了太多判断的结论。
Ok I made a mistake.
The game harness was doing a lot of heavy lifting for both models by preparing optimal options for Jev/Haiku to pick from, given a game state.
If I change the harness to provide only game rules and game state, and only allow Jev to decide the control buttons (left/right/rotate/drop), it pretty much becomes useless.
引用@tdinh_meTetris battle: Jev vs Claude Haiku 4.5
Both compete in REAL TIME.
This means both SPEED and INTELLIGENCE matter.
(Link to try it yourself below) https://t.co/wLsCJWGsii↗
引用@tdinh_meTetris bench: Jev vs. Laya-mlx
I run the Laya model on my MacBook. It's true that the model is really fast (~84ms), but it's also much more stupid.
Losing to Jev 3 out of 3 rounds. https://t.co/LBkcBbi62c↗
Tetris bench: Jev vs. Laya-mlx
I run the Laya model on my MacBook. It's true that the model is really fast (~84ms), but it's also much more stupid.
Losing to Jev 3 out of 3 rounds. https://t.co/LBkcBbi62c
ok, this is a cool project idea:
1. hire a painter
2. post up in an SEC tailgate during rivalry
3. painter creates painting, same-day
4. auction painting to fans
引用@WoodsSavannahI talked to the women who run this tent. They said the artist started live painting at 4 am on Saturday from The Grove. This painting was auctioned and sold for $10,000. https://t.co/wXcMCyy4Z4↗