# Posts on X before Dec 28, 2025, 3:10:52 PM UTC Canonical: https://michaelstolarz.com/x/before/1766934652000-2005295390430740802/ ## On X ### [Dec 28, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2005094198689890475) > Opus, gen with nano banana pro please > \>"That's not a real model — sounds like you're testing me or having a bit of fun." > > It's real, Opus. Web search and see for yourself > \>"Interesting — let me verify this is actually real..." > \>"It... actually worked?" > > Read the chat, it's great [Attached image](/x/media/691504ce56473eaadbd9a2bc5a061e58188f8214bb9d7228c753b77f3fa6584b.jpg) ### [Dec 27, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2004812107842552186) > things are already way ahead EOY 2025 than I thought they'd be at the start of the year. > > given exponentials and all that, seems like this year will even further outpace my already-insane (to the layman) expectations > > will revisit this EOY 2026 Quoted post by Greg Kamradt: ### [Dec 26, 2025 — @GregKamradt](https://x.com/GregKamradt/status/2004689800922853378) > If you’re paying attention, it’s overwhelming ### [Dec 26, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2004405392152187222) > im wondering if we can piggyback off the mental zombification and steer towards actually learning something, and i mean absolutely anything at all, instead of just full pure 100% filler slop > > maybe tung tung sahur can show different types of wood? > > AI cat ones teach about Ikigai? Quoted post by kache: ### [Dec 26, 2025 — @yacineMTB](https://x.com/yacineMTB/status/2004370046244508070) > What's really surprising about how horrible the short form video that is actually zombifiying all of our families is that there is a small group of maybe 100 people max that are responsible for it, and if they all stopped, a great deal of child and elderly suffering would stop ### [Dec 25, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2004190108216705276) > we need to generate more web software that generates react websites which sell how-to-generate python courses which teach you how to generate web software that can generate python websites so that you can generate react SaaS websites so you can charge $9/mo to help people gene... Quoted post by ludwig: ### [Dec 25, 2025 — @ludwigABAP](https://x.com/ludwigABAP/status/2004177597945757698) > More slop, faster. Accelerate slop. Build more things that don’t matter. Ship slop faster. As a matter of fact, don’t even write the slop. Let it be generated. Generate everything. More web. More react. More python. More web software. [x.com/github/status/…]() ### [Dec 24, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2003977398610809097) > i used to make a lot of separate single use apps for my workflows > > now i just make a skill > > i am no longer an anthropic basher regarding their agent supplementations ### [Dec 24, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2003721831296172411) > honestly i couldn't have imagined us being where we are in regards to open models at the end of 2025 with the power level that is [@minimax\_ai]() 's M2.1 > > which means that I will be exponentially more shocked next year, this time. > > lets see if that proves true. retweeting in 1 year. ### [Dec 22, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2003131930716631468) > every day I refresh Groq's model page to see if they added a new model. MiniMax M2? M2.1 now? GLM 4.6? Now 4.7 just released? > > nope. still only offering kimi k2. ### [Dec 22, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2002893369723531396) > this is true because since Q4 2025, with Opus 4.5 / GPT 5.2, a complete normie with 0 coding experience can easily fine tune a model or even make one from scratch > > now imagine if a normie can do that, > > imagine what the top minds of the world that created these models can do now Quoted post by bayes: ### [Dec 21, 2025 — @bayeslord](https://x.com/bayeslord/status/2002562032416272608) > “timelines” don’t matter anymore. we’re in it. the machine is off to the races. ### [Dec 21, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2002792395109527750) > How quickly have we gone from complaining about a model's intelligence to "yeah it's AGI but it's way too slow" > > Intelligence is ~solved soonish; what's next after speed? What happens when there's GPT 6 with Groq/Cerebras speeds? > > Intelligence -\> speed -\> ...financial gain? ### [Dec 21, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2002750814142767348) > omg this is patient 0 btw > i wonder where he is now. someone needs to check in Quoted post by Ariel: ### [Dec 21, 2025 — @ArielKwiat](https://x.com/ArielKwiat/status/2002722668559380510) > LLM psychosis before it was cool [Attached image](/x/media/6dcfa4452f7bfdcbf536468f89f1132050bd5d93ba2143ab927fc6d548054bf5.jpg) ### [Dec 21, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2002704570972328011) > I went back to using GPT 5.2 X-High instead of the new GPT 5.2 Codex X-High > > The Codex version feels much dumber, at least for the work that I am doing. GPT 5.2 -just gets it-, while the Codex variant needs to be handheld althroughout. ### [Dec 21, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2002655756508917912) > The only thing stopping me now from completely dropping Claude Opus 4.5 and going 100% GPT 5.2 Codex X-High is speed > > I've always been one of the "I don't care about speed as long as it's the smartest SOTA model"... > > but after fondling Opus 4.5 for a few days... I miss the speed. ### [Dec 21, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2002560646114840838) > How did this work out last time, ~250 years ago? Quoted post by Helmut Brandstätter MdEP: ### [Dec 19, 2025 — @HBrandstaetter](https://x.com/HBrandstaetter/status/2001945719327043989) > We Europeans have our laws. You will have to learn to acccept them [x.com/undersecpd/sta…]() ### [Dec 20, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2002419453653627194) > plan mode finally in codex btw [Attached image](/x/media/a72088004cd55eda2f5f748e3067f9699be647bc9aa36ed8e67b5a64dac5a32c.png) ### [Dec 20, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2002414390482809073) > Unlike trying to do creative writing with LLMs, coding is something that, for the most part, has a particular answer, if not an exact answer. > > Programming will be "solved" much sooner than creative writing getting to the point where you will preferably read AI-generated fiction. Quoted post by Samuel Albanie 🇬🇧: ### [Dec 20, 2025 — @SamuelAlbanie](https://x.com/SamuelAlbanie/status/2002336809356046697) > update to That Graph [Attached image](/x/media/9b729a68abcfe4a808bfacb8a5e72ce3760355d181db5c7fdc974977a93f0c89.jpg) ### [Dec 20, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2002362392173551648) > Vibe code with opus 4.5 ultrathink > > Fix its tech debt with GPT 5.2 Codex xhigh > > Double check any missed tech debt with GPT 5.2 xhigh ### [Dec 19, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2001928652540530910) > I regret having tasted the speed and niche configurability of Claude Code + Opus 4.5. > > I know GPT 5.2 / GPT 5.2 Codex are smarter (xhigh versions), but damn it are they slow. > > By the time they are done, even though their output is perfect, I forget the direction of development. ### [Dec 18, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2001758927013503292) > when codex cli \<-\> atlas interoperability pls [@ah20im]() [@thsottiaux]() Quoted post by Claude: ### [Dec 18, 2025 — @claudeai](https://x.com/claudeai/status/2001748046175183284) > Using the extension, Claude Code can test code directly in the browser to validate its work. Claude can also see client-side errors via console logs. > > Try it out by running /chrome in the latest version of Claude Code. [Video on X](/x/media/aee70beb4c11a413fdbbc1d88c087832c2cb968b080ca853f6614e2b5ba73366.jpg) ### [Dec 18, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2001584634250068259) > imagine mogging Dia Browser with: > > - a 14 second clip > - of a vibecoded chrome plugin > - it does your entire product's vision > - executes it all flawlessly in a tab Quoted post by Matt Stallone: ### [Dec 15, 2025 — @mattstallone](https://x.com/mattstallone/status/2000702961962418257) > I shipped a feature to [@posthog]() this week, except I don’t work at posthog and posthog didn’t approve it [Video on X](/x/media/6a92a132c9b1ad234452a903dd20b32a86295e28ba2e0b09fdbb4259dabeac76.jpg) ### [Dec 18, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2001541240496763321) > one of my biggest issues with codex has been solved ever since gpt-5.2 xhigh. > > before: first prompts produced the highest quality results, but continuations felt rushed, guessed, non-important > > now: every prompt feels like the first prompt experience. i rarely /new anymore. ### [Dec 18, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2001538270136778928) > nvim and codex is all you need to be cool these days > even cooler if you use codex headless > > feels like the same cool as using nvim in the 2010s and having a super riced out config and obscure plugins no one knows about but are secretly productivity gems > > cool \!= money though Quoted post by alex fazio: ### [Dec 17, 2025 — @alxfazio](https://x.com/alxfazio/status/2001302744963936684) > preferring nvim over vscode > \> cool > preferring cc over cursor > \> not cool > explain this. you can't ### [Dec 18, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2001535174992187888) > how has no one created a degradation benchmark yet? > > check how it performs at different times throughout 24 hours? on weekends? "lazy/holiday EU months"? > > I know there was a paper released roughly about this in 2023 but we're due for a way more tried and tested version now Quoted post by Taelin: ### [Dec 17, 2025 — @VictorTaelin](https://x.com/VictorTaelin/status/2001431627231080867) > No there's no way this is the Opus 4.5 I used a few days ago. > > This thing is brain dead. It is completely moronic. > > I can't accept this is the same model. ### [Dec 18, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2001533645274910776) > and now, tomorrow they're releasing Opus 5.1 > > it's just going to be called Codex Max 5.2 X-High > > and yeah, still slower, but again - you're trading off speed for dozens of IQ points. Quoted post by Michael: ### [Dec 17, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2001170208103559331) > you can use claude opus 5 RIGHT NOW. > > no, not opus 4.5. > > opus 5. > > the only downside is that it's much slower. so if you're fine with trading speed for intelligence, go ahead and use opus 5. > > they named it weird this time though, "gpt 5.2 xhigh", and only available on codex cli ### [Dec 18, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2001532951318917581) > crazy how I can bet on literally anything on gamba sites ("will jimmy steal sarah's lunch from the breakroom again this saturday?") > > but i can't somehow bet/invest on xAI/SpaceX lmao > > [@elonmusk]() isn't X the everything app? can one of those everythings be investing in non-public? Quoted post by @levelsio: ### [Dec 18, 2025 — @levelsio](https://x.com/levelsio/status/2001448146270920986) > [@ChrisJBakke]() Yes please [@elonmusk]() let me invest in xAI ### [Dec 17, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2001170208103559331) > you can use claude opus 5 RIGHT NOW. > > no, not opus 4.5. > > opus 5. > > the only downside is that it's much slower. so if you're fine with trading speed for intelligence, go ahead and use opus 5. > > they named it weird this time though, "gpt 5.2 xhigh", and only available on codex cli ### [Dec 17, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2001117658784395644) > the slow boil death of the US is being brought on by people like bernie and warren > > \>warren "anti-monopoly" crusade against irobot being bought by an american company (amazon) > \>irobot now not only gone bankrupt but was bought up by china and gutted for data > > this smells the same. [https://t.co/zKr3ekW9iU]() Quoted post by Sen. Bernie Sanders: ### [Dec 16, 2025 — @SenSanders](https://x.com/SenSanders/status/2001057004370948131) > I will be pushing for a moratorium on the construction of data centers that are powering the unregulated sprint to develop \& deploy AI. > > The moratorium will give democracy a chance to catch up, and ensure that the benefits of technology work for all of us, not just the 1%. [Video on X](/x/media/78a55c79e781405d61f49021cbe38bf195f3a7e1d428b4c56455b186b1ee9cd3.jpg) ### [Dec 17, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2001113097457778716) > crazy that we can "invest" in prediction markets like "is jerry going to microwave fish in the break room again on thursday dec 18 2025" > > yet we can't invest into obvious long horizon wins like the boring company or space x without doing mental gymnastics "indirect investing" [https://t.co/rzNKHaKyIc]() Quoted post by Elon Musk: ### [Dec 16, 2025 — @elonmusk](https://x.com/elonmusk/status/2001040385272111201) > The Boring Company is so underrated lol [x.com/boringcompany/…]() ### [Dec 17, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2001107461986771262) > all of those ragebait ad games where you have to outline the shape without going over the same path twice are finally going to be useful for designers using this Quoted post by Lukas Ziegler: ### [Dec 16, 2025 — @lukas_m_ziegler](https://x.com/lukas_m_ziegler/status/2000990494575214663) > Supportless, free-form 3D printing\! ⛓️ > > By using low-melting-point Fields' metal and shaping it through surface tension, this process can create highly conductive 3D structures in a single pass, no supports, no post-processing. > > Pretty interesting way of thinking about additive manufacturing, especially for electronics and functional metal parts. > > Just wow. I only know this is from Singapore, anyone has more information about them? > > ~~ > > ♻️ Join the weekly robotics newsletter, and never miss any news → [ziegler.substack.com]() [Video on X](/x/media/149099adc548fa117b6a8aa6615ebdc602bc30abce0f7872ad67f42f2cf5f70b.jpg) ### [Dec 17, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2001081921489789102) > openai's new image gen is a quality upgrade over itself, but definitely not better than google's banana pro image gen. > > the yellow tint is gone, but notice the amount of fingers my left hand has, it makes my eyes funky. > > banana pro result and prompt used for both in comments [Attached image](/x/media/fbafef1051b3fa222dcecad3b6ca1b88bcc1cdc549f1fe7192f5a4a602a00486.jpg) ### [Dec 16, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2000728030801530958) > "they danced on charlie kirk's grave, why not dance on reiner's?" > > because you shouldn't be like them, moron ### [Dec 14, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2000106377939972457) > we're already in the "what are the \_ regulations in space? can you do anything there?" phase > > next phase: elon or bezos (probably elon\!) figures out how to create a country in outer space (or buy it i guess? satellite-level real estate?) > > then says "come dock and arrest me then." Quoted post by terminally onλine εngineer: ### [Dec 13, 2025 — @tekbog](https://x.com/tekbog/status/1999913348595662885) > what are the compute regulations in space? can u do anything there? i mean who could possibly find out [x.com/PhilipJohnston…]() ### [Dec 14, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2000104883199127905) > one of my guilty displeasures is spy movies and how often unrealistic or improbable things happen > > "don't touch your ear when spying" is probably part of day 1 orientation at spy school > > yeah i get it, it's roller coaster actionslop entertainment, not a documentary. but still Quoted post by James Bond: ### [Dec 13, 2025 — @007](https://x.com/007/status/1999826691565236512) > Stop. Touching. Your ear. [Video on X](/x/media/6d62132240f8031504e009abc23a8b36a9eae4800ef4288f107e7d251b29b156.jpg) ### [Dec 14, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/2000027091300180447) > it's crazy to me that it's the complete opposite for me > > i use opus 4.5 for speed and when it inevitably starts making mistakes/it's time to clean up its spaghetti code, gpt 5.2 xhigh comes in to save the day (albeit slowly) Quoted post by Nathan Lambert: ### [Dec 13, 2025 — @natolambert](https://x.com/natolambert/status/1999868051915985138) > Too many cases of starting something in Codex that you think is going great then Opus needs to save the day. ### [Dec 13, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1999775455814303767) > not to mention they now have Opus 4.5 \& GPT 5.2 xhigh to help them speed things up > > the wheel spins faster. Quoted post by Chris: ### [Dec 13, 2025 — @ChrisGPT](https://x.com/ChrisGPT/status/1999748455124611378) > Hard to imagine how much better Opus models can get with 23B$ worth of tpus from Google over the next 2 years ### [Dec 13, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1999725024669012230) > i always wondered if someone somewhere at least once suffered an attack from this Quoted post by Sawyer Hood: ### [Dec 13, 2025 — @sawyerhood](https://x.com/sawyerhood/status/1999648786885542367) > claude code needs an epilepsy warning [Video on X](/x/media/b126dfc2e8ef8cc4e8a15d506a231a911473ee5abeafd71fdbcdafa88fd64a0d.jpg) ### [Dec 13, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1999653618660913416) > gpt 5.2 xhigh on codex went nearly 3h last night > > the result was perfectly what I wanted btw Quoted post by Jon Hernandez: ### [Dec 12, 2025 — @JonhernandezIA](https://x.com/JonhernandezIA/status/1999542515750310071) > WHAT THE ACTUAL F\*CK?? > > maybe they went too far with pro? 😂 [Attached image](/x/media/638a8308a5fe26c58d717209e77635a977ff7deb03ebf69a9544e4aa347ea09a.jpg) ### [Dec 13, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1999653368906891277) > And just like that, I never have to worry about resetting/manually compacting when using GPT 5.2. > > They hit it out the park, as usual. > > Better long context than the Anthropic models, and that's saying something. ### [Dec 13, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1999649626971091014) > i just dont understand the play here. this is how his time is best spent? what is the end goal? is this his childhood dream he's now pursuing because he can just throw millions at it and force it to happen? > > no way is this financially sound right Quoted post by Dom Lucre \| Breaker of Narratives: ### [Dec 12, 2025 — @dom_lucre](https://x.com/dom_lucre/status/1999629992603582951) > 🔥🚨DEVELOPING: Hollywood actor Will Smith is continuing his attempt to make a comeback to rap music by collaborating with influencers. Will teamed up with Adam W as he tries to appeal to a younger audience. [Video on X](/x/media/087776453ab2ff4b29a203a9dbae890a7655c07cb4f544e471dffc4ea1badf85.jpg) ### [Dec 12, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1999484942678655468) > WOAH idk how I missed this crazy detail. this alone takes things to the next level imo. this was always openai's weakness Quoted post by Peter Steinberger 🦞: ### [Dec 12, 2025 — @steipete](https://x.com/steipete/status/1999444593197408642) > The biggest win for GPT 5.2 is August 2025 as knowledge cut-off date. That's almost a full year more. > (5.1 was mid 2024, Opus is at March '25) > > This + the close timing really pushes the theory that 5.2 should have been 5.0 but sth didn’t work out back then. ### [Dec 12, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1999482463450726560) > lmao remember they were always tagging disney saying all this pr slop of "did u know about this ai content using disney IP, i will never watch disney again if you condone this" > > they must have seen it and said damn, if you cant beat em, join em Quoted post by Angel 🌼: ### [Dec 12, 2025 — @Angaisb_](https://x.com/Angaisb_/status/1999466419642392933) > Anti-AI people are going crazy after the OpenAI-Disney deal ### [Dec 12, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1999472816211788208) > gpt 5.2 is the first model that just... gets it. iykyk. ### [Dec 12, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1999470137917997195) > imagine the smell of the people who think saying "imagine the smell" is racist ### [Dec 12, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1999464840642109697) > PLEASE CHATGPT > > the only reason why I am not 100% daily driving [@OpenAI]() chatgpt atlas as my browser is because it's missing one crucial component to my workflow > > spaces. once chatgpt atlas has spaces, it's game over. [Attached image](/x/media/be580a080bf2351d7da55938d114f4e50ffcef3ea7909cc3f0b295904a353dff.jpg) ### [Dec 10, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1998592428455108867) > 100%. one of my favorite hobbies is watching old programmers with 50+ years of experience in the industry, telling me why they're such an amazing programmer, "i write my code in a way where i can be sure someone in 10 years from now can understand and implement further" > > sir... [https://t.co/syN17OfuOk]() Quoted post by terminally onλine εngineer: ### [Dec 9, 2025 — @tekbog](https://x.com/tekbog/status/1998332419502682322) > it’s honestly incredible to see software decline in real time these last couple of years > > i think there are different factors contributing to this > > nobody cares about the actual craft anymore, which is the consequence from: > 1. leetcode interviews and faang interview style adoption - dsa is good but it doesn’t capture real knowledge on building systems > > 2. constant layoffs and ship jumping with no consequences - ie promo driven culture, ship and dip, no one is held accountable because nobody stays in charge of the product/system long enough to even have the full domain knowledge > > 3. non technical PMs in charge of product, services and pushing engineering to just ship with little understanding of the space, with only regard for themselves, because they too want that promotion and who cares if prod is broken > > 4. software engineering is not taking seriously - call it development, programming, whatever but nobody wants to have ownership and people dont care, just try to make the devs ship and the devs try to shit out a bunch of awful stuff to stay employed > > 5. with the addition of ML/AI into the picture we have a lot of brilliant folk who just wanna work in the AI space, which ofc is interesting but so is software engineering as a whole, a lot of challenging problems everywhere and AI is part of swe id say, we have people trying to ship AGI unable to run a crud app > > 6. the death of excellence - this feels like a global problem everywhere you go and look at, the decaying state of things, lack of domain knowledge or willingness to learn - a global cultural phenomenon? maybe i need to touch some grass [Attached image](/x/media/b82415f23f6986a898d5d142e43b7fc82bf8890057a36e42188ae50e2f2155d8.jpg) ### [Dec 9, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1998259708143231408) > it feels like this because 99.9% of the time it's either ragebait or they are so incredibly wrong you would need way more than a tweet to even nudge them towards the correct direction Quoted post by A.I.Warper: ### [Dec 8, 2025 — @AIWarper](https://x.com/AIWarper/status/1998130675363536953) > I’m at the stage in my life where, if I’m about to reply to someone who’s completely wrong about a subject I’m literally an SME in, I stop immediately and close the app. > > I suspect a lot of us do this in modern internet culture, tbh. ### [Dec 8, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1998116623585255759) > we're really shifting from a world of man vs. man into man vs. self > > ironically those that would have been the most ferocious warriors of the past are instead today's guiding light in the ever-deepening tunnel of self despair ### [Dec 8, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1997920052507824606) > if you talk to your AI as if it's a terrible programmer, correcting its slightest mistakes with demeaning namecalling and ridicule, > > it will continue to play the role you've just given it ### [Dec 8, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1997919722923585750) > whether you think the AI can or cannot, > you're right. ### [Dec 8, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1997887470684557380) > recently went from 1g (yes, 1,000mg) of caffeine to 200mg > > used to take it (200mg) in the morning, 11am, 1pm, 4pm. > > now only in the morning after waking cortisol spike is over > > cleaner focus. the previous dosage added "feel-good" but no performance increase ### [Dec 7, 2025 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1997620228998725802) > this is true for the web apps but completely the opposite for the APIs. I have no idea why > > I can't imagine doing mathematical work with any claude model, the only ones I trust are codex 5.1 xhigh and gemini 3 pro. > > everyone else is guessing trash, literally "eh, maybe this?..." Quoted post by Code With Antonio: ### [Dec 7, 2025 — @codewithantonio](https://x.com/codewithantonio/status/1997619159920701725) > i've been a claude user for a while now, and decided to give chatgpt a chance again 🙃 > > it is borderline unusable, and the way it tailors it's answers to what it thinks you want to hear is almost laughable > > i don't know if it is because those are the most positively-marked answers, or if it is being too cautious not to offend you > > you literally cannot trust it with a grocery list, fitness advice, or god forbid anything that includes numbers [Older posts](/x/before/1765104751000-1997620228998725802/index.md) [More from @michaelstolarz](https://x.com/michaelstolarz) Saved from X: 2026-09-07T03:42:59.938Z