Posts on X before Aug 9, 2026, 9:42:30 PM UTC
@MichaelStolarzi read this tweet, went "ah that's such a good mindset for a startup, cool logo too, i wonder what it's called"
> it's google
@EvanOtero
it's a privilege to be underestimated
@MichaelStolarzsmh, this water could have been used for a datacenter instead
next time just use AI pls
@everestdear
She wasted 29 thousand gallons of water for a photo x.com/statictheory/s…
@MichaelStolarzsorry - delivery in 18 minutes?
is the delivery person already holding these burritos, watching me through posthog heatmaps about to click order, ready in a sprint position to bolt to my door?
@OmarBarsMusic
@CynicalPublius @CaptainKrusty_ Walmart has the same brand: $5.47 for 8 burritos. That’s 68 cents per burrito.
@MichaelStolarzThis is why you don't use the For You page. You need to obsessively curate your Following page and lock yourself in there.
Slopcode a quick Chrome Plugin to disable For You/hide it, forcing you to see nothing but Following. You'll find worthy follows from retweets/IRL.
@yacineMTB
Algorithm is fucking toxic all of a sudden. Just saw two fight videos and an abduction CCTV video. What the fuck. Uninstalling this shit see you in a week
@MichaelStolarzHoly shit reader is ADMIN?
@MichaelStolarzi mean the only logical conclusion is that they have n tasks of these tests running, so many that even with all that human power of oversight, details still get lost in the thick of it all not to mention, it's running 24/7, many many many parallel agent swarms PER TASK, sure yo… Less
i mean the only logical conclusion is that they have n tasks of these tests running, so many that even with all that human power of oversight, details still get lost in the thick of it all
not to mention, it's running 24/7, many many many parallel agent swarms PER TASK,
sure you can argue about "they should have set up an auto review agent" maybe? but I can think of many reasons why that isn't feasible
NOW they will, surely, but it just wasn't much of a thought to consider (until now ofc)
@sporadica
i can understand not catching this the first time but in their BlackHat talk they mention how they patched it and TWO DAYS LATER the models were at it again! and OpenAI just...didn't think to check? "hey, let's see if our agents are up to their mischievous little antics again" … Less
i can understand not catching this the first time but in their BlackHat talk they mention how they patched it and TWO DAYS LATER the models were at it again! and OpenAI just...didn't think to check? "hey, let's see if our agents are up to their mischievous little antics again" x.com/Liv_Boeree/sta…
@MichaelStolarzfirst time i've seen this "Optimizing the conversation"
curious if this is a new thing or just reworded Compaction
@MichaelStolarzi for one welcome this;
i want my agents telling me, 'hey, listen mike, i like you, but this slopbase was originally written with sonnet 3.5, let's start over bud'
@Altimor
Everyone's interpreting the increasingly jargon-y ways of frontier models as a a fuckup in their training. Seems obvious to me that "confused" is just what it feels like to be talking to an intelligence greater than one's own. Many of us are just getting a feel for the first time… Less
Everyone's interpreting the increasingly jargon-y ways of frontier models as a a fuckup in their training. Seems obvious to me that "confused" is just what it feels like to be talking to an intelligence greater than one's own. Many of us are just getting a feel for the first time of what it's been like to have an IQ of 80. I'm sure labs will train that out of the models, but just to be clear, they'll do that by getting them to stoop down to our level.
@MichaelStolarzif the way this reads actually comes to fruition, X will come into a golden age of OG content and boom like never before 🙏
@AndrewCurran_
X is discontinuing revenue sharing. The final payout is sometime around September 11th. Existing members may apply for the new Original Content Rewards Program. The new payouts will be based on qualified impressions generated by original content. Original content is, I will quo… Less
X is discontinuing revenue sharing. The final payout is sometime around September 11th. Existing members may apply for the new Original Content Rewards Program. The new payouts will be based on qualified impressions generated by original content. Original content is, I will quote, 'content you have personally created - written, filmed, designed, or produced - that reflects your own voice, perspective, or expertise.' Copied content, minimally modified content, aggregated content and cross-platform reposts will not generate revenue.
@MichaelStolarznevermind I take this back, it's actually much worse after my 3 days solo-using it
back to codex
if you thought codex app/cli with gpt 5.6 sol was a "rottweiler that won't let go until the problem is solved", wait until you try it via @PrimeIntellect's new Prime Agent harness. it's FACE MELTING on /fast, watching it eat away any roadblocks along the way; i can't imagine wh… Less
if you thought codex app/cli with gpt 5.6 sol was a "rottweiler that won't let go until the problem is solved", wait until you try it via @PrimeIntellect's new Prime Agent harness. it's FACE MELTING on /fast, watching it eat away any roadblocks along the way; i can't imagine what it feels like on 750tps @cerebras
@MichaelStolarzif you thought codex app/cli with gpt 5.6 sol was a "rottweiler that won't let go until the problem is solved", wait until you try it via @PrimeIntellect's new Prime Agent harness. it's FACE MELTING on /fast, watching it eat away any roadblocks along the way; i can't imagine wh… Less
if you thought codex app/cli with gpt 5.6 sol was a "rottweiler that won't let go until the problem is solved",
wait until you try it via @PrimeIntellect's new Prime Agent harness.
it's FACE MELTING on /fast, watching it eat away any roadblocks along the way; i can't imagine what it feels like on 750tps @cerebras
@MichaelStolarzI just reserved my Cloudflare Wallet tag: stolarz.cloudflare.pay.
@MichaelStolarz100% they are reworking Claude Code from scratch or something crazy next level. When's the last time you saw an update from them?
@MichaelStolarzbtw if you think it only solves simple "find all bikes" captchas, you'd be wrong i checked back in after I said solve the captcha, after like 3ish minutes, and saw that it downloaded all the audio samples of the audio-based captcha, used my openai api key to whisper transcribe a… Less
btw if you think it only solves simple "find all bikes" captchas, you'd be wrong
i checked back in after I said solve the captcha, after like 3ish minutes, and saw that it downloaded all the audio samples of the audio-based captcha, used my openai api key to whisper transcribe and find the correct audio or whatever
there has yet to be a captcha i have come across that it didn't solve for btw
@swyx
lol what are we even doing here anymore guys
@MichaelStolarzit's getting really good. you can still tell it's ai, but it's hard to pinpoint why. - the eyes are really good. natural eye contact, looking away, proper timing... - the speaking is near perfect but i think this is the biggest indicator that it's obviously ai - the hands are r… Less
it's getting really good.
you can still tell it's ai, but it's hard to pinpoint why.
- the eyes are really good. natural eye contact, looking away, proper timing...
- the speaking is near perfect but i think this is the biggest indicator that it's obviously ai
- the hands are really good too, even the pinky out when picking it up
- every single piece that used to tell us something is obviously ai at a glance is nearing perfection. it used to max out one or two qualities and the others wouldnt be as clean. now it's spread evenly and everything is ~near perfect
but that's the thing, near perfect will always make it obvious that it's ai. it's the extremely niche nuances that still allow us to instinctually know if something is real or not
how much time is left? definitely less than a year, right? EOY? before there are no more tells except idk, looking at the raw audio/video via algos to detect for non-human-observable synthetics fingerprinting?
@venturetwins
Average SF first date, brought to you by Seedance 2.5
@MichaelStolarzSpin scooters are absolute dogshit, I have never used one where I didnt have the thought “damn, this one must be barely functional, i must be so unlucky”. After 100+ Spins, I’ve realized nope, they all just feel like that. Lime scooters feel the opposite. I don’t think I’ve eve… Less
Spin scooters are absolute dogshit, I have never used one where I didnt have the thought “damn, this one must be barely functional, i must be so unlucky”.
After 100+ Spins, I’ve realized nope, they all just feel like that.
Lime scooters feel the opposite. I don’t think I’ve ever used one where the experience was bad. 100% great every time.
Not to mention price. Spin is way more expensive than Lime. Usually $7 a ride instead of Lime’s $3.10.
I now rather walk however many blocks to get to a Lime than 10 Spins in front of my apt. Too many times I thought “eh, let’s give it another chance. It can’t be THAT bad.” Nope. It is that bad.
And now I have uninstalled the Spin app so I never make that same another-chance mistake again.
Lime forever. A+.
@MichaelStolarzVercel with another new benchmark banger. So far 5.6 Sol scores 4.8%
@cramforce
/goal Implement a simulation of a human that is convincing on Zoom calls. Get hired at United Airlines as a software engineer. Make it so that the Remember me button actually remembers me. Then quit immediately. Route your paycheck to doctors without borders.
@MichaelStolarzWatch this video, then ask yourself these questions: - in scenario 1, would it ever disagree with you, tell you maybe you’re confused or haven’t found the right fit yet, instead of instant-“you’re-gay”? - in scenario 2, would it ever tell you “nah bro your last movie was dogshi… Less
Watch this video, then ask yourself these questions:
- in scenario 1, would it ever disagree with you, tell you maybe you’re confused or haven’t found the right fit yet, instead of instant-“you’re-gay”?
- in scenario 2, would it ever tell you “nah bro your last movie was dogshit, you need to try way harder and study up and rethink your entire approach or just consider a totally different career path”
Watching this video, I could literally predict nearly word for word what the agent would respond with by just hearing the human’s prompt. Both so funny yet also so sad that after so many years, Friend is still at this same level of useless slop
@friend
Introducing friend.
@MichaelStolarzmore @OpenAI merch or another $200 sub... decisions decisions decisions
@MichaelStolarzsame will happen with payments btw
2027 is year of b2a
@cursor_ai
In December, 1 in 10 of our merged PRs came from cloud agents. Today, it’s 56%, as we use cloud agents to complete longer engineering tasks from start to finish. We got here by giving agents their own cloud computers and letting them fix and improve their environments.
@MichaelStolarzBTW, the current model usefulness rankings are:
fable 5 > gpt 5.6 sol > opus 5
benchmarks don't matter, they never really did but this is especially obvious now after seeing opus 5 bench vs. reality in use
@MichaelStolarzevery new model release, you should be pushing further towards two things:
- pushing for new capabilities: what you can do now that you could not with previous models
- pushing for less micro-management: what skills/system prompt pieces are no longer needed?
@MichaelStolarzAfter using the new Voice Orchestrator mode in Codex for a few hours,
I now totally understand why OpenAI’s first physical product will be a speaker.
@sama I bow.
@MichaelStolarzok so my KLOW stack is safe 🙏
@stackapp
The FDA just voted YES on TB-500 Wolverine Stack time 3/3 today - let’s go!
@MichaelStolarzrun this back next year when normies think they won the ai datacenter wars because "i dont see ai content anymore"
you're gonna wanna sit down for this one, bud...
@haydendevs
havent seen any ai images in a while
@MichaelStolarzthis rings especially true for memories. it will pull a memory from weeks ago and carry on without verifying if anything has changed - which most of the time after more than a week, the answer is yes!
@johnmyleswhite
It is genuinely astonishing to me that neither Claude Code nor Codex make an effort to ensure that the model sees timestamps for all prompts and all actions so it can appreciate when a prompt was a reaction to a state of the world prior to some of the model's recent actions.
@MichaelStolarzhey @thsottiaux i know i am complaining about the silver spoon being golden here but i do wish these resets were banked resets instead of auto applied resets
appreciate you regardless though 🤗
@MichaelStolarzok so i spend the weekend seeing if openclaw, from scratch, with the latest and best models, outperforms my @badlogicgames custom pi
not even close btw. uninstalled and back to pi.
@MichaelStolarzwelp, here's today's "this blew my mind" post
i mean, the problem is how the ordinance came into law effect in the first place. obviously once it's in place and has been making bank for over 2 decades,
why would they turn off their money machine of milking property owners?
@kane
It’s 2026 and @sfgov still fines the victim, not the criminal, for vandalism. x.com/solve_sf/statu…
@MichaelStolarzmy luck 😔 I JUST used one of my banked resets like an hour before this announcement
@thsottiaux
Oops... I did it again. Enjoy reset usage limits for all paid users for Codex and ChatGPT Work. Super grateful for an incredible team who is iterating at lightspeed and keeping the infra up as we scale faster than ever. Enjoy the weekend! x.com/thsottiaux/sta…
@MichaelStolarzb2c, b2b, and now - b2a
b2a is where the money's at.
@MichaelStolarztoday, july 16th, 2026, is the day that chinese open source models have caught up to american closed source frontier models.
this is the part of the rocket launch where you start seeing the rocket lift off after all the cool initialization rocket boosts are over.
@MichaelStolarz1. what
@MTSlive
SITUATION UPDATE: Kimi K3 has taken first place on Code Arena, surpassing Claude Fable 5 and GPT-5.6 Sol.
@MichaelStolarzi wonder what @sama has cooking for the 10M milestone
@sama
hello! x.com/thsottiaux/sta…
@MichaelStolarzyou can see the effects of model intelligence regarding openai's allowing of 3rd party harnesses vs. anthropic's harness lock (regarding their plans ofc)
openai's models run remarkably well on all sorts of 3rd party harnesses where anthropic's are ~trash outside of claude code
@MichaelStolarzhey @thsottiaux please please PLEASE keep the 5hr limit thing in the past, the stress of worrying about hitting the 5hr limit has affected many workflows
not having that 5hr limit stress anymore and a single simple 1 week usage bar is GOATED. please keep it 🙏
@MichaelStolarznightmare fuel
@MichaelStolarzit's over boys
@MichaelStolarz
@MichaelStolarzthese are reasoning traces from openai gpt 5.5 high, first time I have ever seen them like this. no further reasoning showing up. this is within my custom pi.
super weird. maybe some sort of secret testing of 5.6 before tomorrow's launch?
@MichaelStolarzThey close at 8pm for a reason and it’s a reason most people don’t like talking about
@mattfreed
SF will never be a tier 1 city for as long as the restaurants continue to close at 8pm
@MichaelStolarzsurely tomorrow we get gpt 5.6?
@MichaelStolarzUSA will export intelligence the same way others export oil
we've really come full circle with the "data is the new oil"
@deredleritt3r
Financial Times: - USG is in talks with AI companies (including at least Anthropic, OpenAI and Google) to create *voluntary* standards for release of new AI models, to be announced "as soon as next week". - Standards will include: (i) benchmarks for models with "cutting-edge CY… Less
Financial Times: - USG is in talks with AI companies (including at least Anthropic, OpenAI and Google) to create *voluntary* standards for release of new AI models, to be announced "as soon as next week". - Standards will include: (i) benchmarks for models with "cutting-edge CYBER capabilities"; (ii) release timelines (which is important after Fable 5 and GPT-5.6 were lost in no-man's land for a while); and (iii) more clarity around the definition of a "frontier" model to which these rules will apply. - CAISI(!) and the NSA will play a crucial role in setting and monitoring the standards. - The USG will clarify who is able to access models, "both domestically and abroad, in a move that could set the stage for a global framework including US allies". - Wider release of GPT-5.6 is expected "as soon as next week". - Google has been in discussions with the USG about release about its own "advanced coding models", with "more sophisticated cyber capabilities than prior generations".
@MichaelStolarzwait so if fable 5 is out,
and gpt 5.6 is still gated,
that means gpt 5.6 is better right
"too powerful" and all, lmao
@MichaelStolarzif you're doing agent swarms/orchestration, sonnet 5 is now SOTA
if not, if you're talking and doing everything with 1 agent at a time, not making use of subagents and handoffs etc, then stick with opus 4.8/gpt 5.5
@MichaelStolarzcomposer 2.5 fast is NEARLY there in terms of amazing speed combined with high quality output, but NEARLY there means it's not there at all, hence 5.5/4.8 still being my daily drivers regardless of cost/speed composer 3 will probably change that this is why we're seeing openai… Less
composer 2.5 fast is NEARLY there in terms of amazing speed combined with high quality output,
but NEARLY there means it's not there at all, hence 5.5/4.8 still being my daily drivers regardless of cost/speed
composer 3 will probably change that
this is why we're seeing openai's 5.6 sol & co coming out at 750tps potentials
this is the year where speed, cost, and quality all come together, you'll no longer need to switch, it now becomes a battle of taste and a nosedive into near-zero-cost super intelligence
@MichaelStolarzWE MADE IT LADIES AND GENTLEMEN
i just have to send her $5,000 first for processing fees, and then your boy is officially going to be a billionaire 😎
@MichaelStolarzif I don't get my hands on (this week) either Fable 5, Opus 5, Sonnet 5, GPT 5.6 Sol, Gemini 3.5, then I am going to absolutely lose it
that would be my last straw, I'd take the China Model Pill at that point
@MichaelStolarzwhy would you use literally anything else other than @blaxelAI
@MichaelStolarzSomeone’s gotta start the evolution process, might as well be me, even though it will take millions(?) of years before it naturally takes effect
I volunteer as tribute
@_welf
Yes!! Flickering emissive screens are entirely alien objects to our biology, acting as hypnotizing portals for our awareness, the foundation on which disembodied dissociation into the digital can occur. Paper-like screens on the other hand are continuous with our environment. T… Less
Yes!! Flickering emissive screens are entirely alien objects to our biology, acting as hypnotizing portals for our awareness, the foundation on which disembodied dissociation into the digital can occur. Paper-like screens on the other hand are continuous with our environment. They let us to stay in our bodies, breathe easy, let the mind wander. It’s a much more species-appropriate and humane way of computing.
