# Posts on X before May 10, 2024, 2:29:38 PM UTC Canonical: https://michaelstolarz.com/x/before/1715351378000-1788939459435929615/ ## On X ### [May 10, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1788932738571251790) > let's chug 2 red bulls, a litre of red horse beer, and code until sunrise. wish me luck 🫡 ### [May 10, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1788802495369883959) > anyone who tells you to "just put it all in the context window, it will be better" doesn't know how context length works for LLMs nor have they tested it out extensively enough to notice the subtle retardation that happens (well, first subtle, then it's very obvious as you continue to add context) > > don't get me wrong; there's a sweet spot. obviously if you put too little context, the result will be worse than if you put just the right amount of context; but past that, the more you put in, aka the more garbage you put in after what's immediately necessary, the worse your end results will end up being ### [May 10, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1788765412626403787) > currently no one knows what the l stands for, here are my guesses, in order from most probable to least probable: > > - gpt-4 lite > - gpt-4 large > - gpt-4 local Quoted post by Rasmus: ### [May 9, 2024 — @synquid](https://x.com/synquid/status/1788653305716969875) > New OAI models: gpt-4l, gpt-4l-auto, gpt-4-auto > > It’s over? ### [May 9, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1788548781992214865) > dont forget slop stems from the term feed, as in we are seeing an increase of AI slop in our feed > > we're complaining little piglets asking for better rations Quoted post by kache: ### [May 9, 2024 — @yacineMTB](https://x.com/yacineMTB/status/1788540032141099457) > very cool that we're turning slop into a technical term ### [May 9, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1788535628675834261) > shitposting protip: get a mechanical keyboard (current fav is blue switches) - it makes posting so aesthetic, even if you're typing a bunch of autistic analysis ### [May 8, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1788225566484861425) > Yeah sure [@yacineMTB]() uses c but does he use HolyC? checkmate. Quoted post by tetsuo: ### [May 8, 2024 — @tetsuoai](https://x.com/tetsuoai/status/1788034738105807252) > The late Terrence Andrew Davis explains the differences between HolyC and C++. He discusses HolyC's unique approach to header files, task management, and symbol tables in Temple OS. [Video on X](/x/media/c0a8d0ba0db81adf71613826465ce2a6403f2c9c9946ace064898e7dee8b76e1.jpg) ### [May 8, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1788165103541715370) > agreed. apple is primarily known for its security and privacy. scoff all you want, android users/apple haters; there's no better mainstream device that prioritizes security other than apple products. > > no, your fringe modular rooted android with sys component replacements doesn't count. Quoted post by Matthew Berman: ### [May 8, 2024 — @MatthewBerman](https://x.com/MatthewBerman/status/1788086729683153076) > The top comment on the hacker news post about the new Apple M4 chip is spot on. > > Long Apple. [Attached image](/x/media/432c4c09bab091c3ad50e04359201d7c6d0a73fe3ad3200e8539dc3760bda904.jpg) ### [May 8, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1788163334958899277) > the problem I have with these takes is that... what other choice do you have? you need to use AI in your business? then why wouldn't you use the SOTA choice? which will only get better? > > take perplexity for example. yeah, they're predictably going to face competition with openai's new search that they're releasing soon. > > but... so what? yeah, I guess they could have used... their own self hosted models? gone a different business route altogether? not done perplexity in the first place? pivoted earlier? > > one version is you make a lot of money with the possibility of being replicated by the service you are benefiting from. > > the other version is... you don't make money? you choose something else that... doesn't have to do with AI? [Quoted post unavailable](https://x.com/i/status/1788148550527783408) ### [May 8, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1788162263578488924) > unironically though we're probably less than 20 years from the robotic early versions shown in westworld (not the flesh ones... yet) Quoted post by Smoke-away: ### [May 8, 2024 — @SmokeAwayyy](https://x.com/SmokeAwayyy/status/1788051192565969050) > China is already on Westworld Season 1. > [pic.x.com/0WKgoI4Dot]() [Video on X](/x/media/6118ebb2530ba79ed500cef0133624236d98f3c5d24ee8ed7c4ad513a571650f.jpg) ### [May 8, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1788161244412940768) > OK, I've seen this type of behavior A LOT recently from ALL the llama 3 models. here's what i think is going on: > > first, there was no need for it to make up a word. it could have just used... preface. "I must preface" would have worked just fine > > from heavy personal experience with all the models, it seems like they always try to show off their "humanity" within the first part and last part of their responses. > > it often starts off with something cringey along the lines of "Brilliant\! What an interesting take." - and then it gives you the actual answer to your query / conversation > > similarly, it ends the same way as well - many times, I've had it end with a response like "so there you have it\! Thanks for listening (or, READING - heh) and I hope I answered your question there." > > yes, this satisfies those that begged for a more humane talking companion but it also seems to introduce these creative(?) elements as seen here: making up words, catching itself as a human who made a speech mistake would, etc. > > I think it's a good step towards a less robotic, more enjoyable conversational system, with 1 caveat; I want to be able to turn this type of behavior off. hopefully one can do so through some additional system prompt rulesets (though personally I haven't reached a point where robotic behavior is favorable, so I haven't tried) Quoted post by Daniel Kaiser: ### [May 8, 2024 — @spectate_or](https://x.com/spectate_or/status/1788102406250664171) > llama3-120b randomly creates a new word "prefaceate" that has 0 search results on google [Attached image](/x/media/f06e0a3b7eba771b3f02c6a4eafa9ac9bbe5c08d79589176ca37e32dadba1f71.png) [Attached image](/x/media/a30fc19b9df0fde2538fd1ac1685bbb669838e732dd986b8a76bf96d1b0bee30.jpg) ### [May 8, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1788151174828614106) > Hey [@elonmusk]() you should allow full markdown syntax compatibilities in X; imagine code-blocks, headings, it will buff articles, make things more interesting to read. > > Literally no downsides. ### [May 7, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1787867382431760893) > bindu is an engagement farmer. it's why she's muted/blocked Quoted post by Spencer Schiff: ### [May 7, 2024 — @SpencerKSchiff](https://x.com/SpencerKSchiff/status/1787866061557113163) > So much of what Bindu tweets is just nonsense. Like, no, there's a 0% chance that any of the gpt2-chatbot variants are GPT-5. Sam explicitly confirmed this a few days ago ### [May 7, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1787773196651180064) > until you can test out 100K+ token context, all these gpt2-chatbot tests are cute but can't show that they're more useful than their currently available iterations > > yeah they will be a bit smarter, but we really need USEFUL 100K+ context windows, not "128K context but becomes braindead at around 10K tokens" ### [May 7, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1787746390745272626) > currently, simping for AI is cool because it's an American company, cool OpenAI / sama lore, etc. > > what happens when the SOTA model owners start being chinese? are we all simping for the best AI models... made by the chinese? > > I know I will. I have no allegiance; only to the best. Quoted post by Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞): ### [May 7, 2024 — @teortaxesTex](https://x.com/teortaxesTex/status/1787720476988453245) > \*tears up\* they grow up so fast [x.com/teortaxesTex/s…]() [Attached image](/x/media/3799c55faa6f5a653af5f99f42bfcfd41a6dd9d2f956b31f3d191d6f3910fc7d.jpg) ### [May 7, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1787728685857931581) > Until it’s literally a pair of glasses with \>24hr battery life that I can wear and use all day and charge while I sleep, all VR/AR aimed at mass adoption / “iPhone moment” is dead on arrival. > > If you want another iPhone moment in the same user space, you need to replace the iPhone itself. Quoted post by Mitesh: ### [May 7, 2024 — @lml_mitesh](https://x.com/lml_mitesh/status/1787727623285862743) > Is it just me, or has the Apple Vision Pro completely disappeared? It was all the rage just a few months ago. ### [May 7, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1787726063470026999) > Who gives a fuck? Buy a new iPhone if you want a new iPhone. This whole thread is commie fucking brain rot circlejerking who can be seen as the most ethical. > > Stop caring what others think and think for yourself. Use your time more effectively elsewhere instead of crying about nonsense. > > “You should use your phone until it falls apart and then buy an old refurbished one” yeah retard THAT’S gonna save the planet for sure 👍 > > On second thought, yeah, go do that. That way I can buy my wife the pink iPhone 16 when it comes out in September instead of having to wait since you idiots will be buying old broken androids instead. Thanks in advance 🤝 Quoted post by doofdoof: ### [May 7, 2024 — @braedd_](https://x.com/braedd_/status/1787708116223897937) > [@Purplessage]() [@SpiritofLenin]() Thanks for the advice\! I'll keep that in mind. > But just another follow up question: will I be seen as unethical if I have a desire to purchase something brand new? I understand buying used/refurb but ig there's something rewarding about getting a brand new thing for myself. ### [May 7, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1787680516361290060) > this is already the case with gpt-4/claude opus; 99% of people don't even know it exists/what the capabilities are. > > of those that have used AI, they've used chatgpt 3.5 a few times and said "oh cool" and forgot about it. > > we forget that we're at the cutting edge here because of tpot and our obsession with literally anything, even a rumor, about AI. [Quoted post unavailable](https://x.com/i/status/1787678701133455537) ### [May 7, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1787664252947623999) > not to mention it's so cool when you fuck up (and most of the time, mentally useful to see what happens) > > imagine your training wheels never coming off? you'd never experience the crashes, falls, pain, that all ultimately led to you learning how to ride a bike Quoted post by kache: ### [May 6, 2024 — @yacineMTB](https://x.com/yacineMTB/status/1787515199299199158) > you should be using unsafe languages. imagine if i used rust - i wouldn't have experienced this kino memory overwrite bug [Attached image](/x/media/5a14986b4d4a006932342a68ff6a40bc99b0a4a08334644aa71107fc2774c24f.png) ### [May 6, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1787449786377056270) > villains are so cringe if you think about it. imagine coming to the conclusion "i'm going to be evil and do bad things", whether intentional or misguided. > > "oh but xyz happened in my childhood so i have to take revenge on abc" lol ok bro yeah your silly little mischief is so cute ### [May 6, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1787281955744370959) > if gpt2-chatbot doesn't have an (actually working) 100K+ context, preferably closer to 1M, then it doesn't matter at all. > > current gpt-4 can't even handle 10K context without losing its mind. hopefully this iteration/upgrade is better than that. ### [May 5, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1787054511472697674) > I wonder if [@beffjezos]() will go down in history as: > > - “the e/acc meme pioneer who also happens to have made a company called Anthropic” > > Or > > - “the creator of Anthropic who also happens to have made the e/acc meme” > > Both have very different meanings towards a lasting legacy; both are still very cool to have on your human life portfolio though, so it’s win win either way it goes for him 🤝 ### [May 4, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1786637636041683411) > keep in mind: we're talking about a gpt wrapper LMAO Quoted post by Jesse Lyu: ### [Jan 11, 2024 — @jesselyu](https://x.com/jesselyu/status/1745555905545228673) > by submitting as app, you submitting all your codes to them. think about it. remember there’s one of the most popular apps you want early days on app store was called ‘flashlight’ now see what happens? apple just incorporated that feature in iOS. so are building apps sustainable to a startup? maybe not. ### [May 4, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1786636775466963120) > no one cares how many degrees you have. they want to jump into a project that you made from scratch and test it out, seeing your thought process during development. see your commits. your flaws. design choices. ### [May 3, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1786335072397717629) > I paid for the highest tier X Premium+ or whatever it’s called. Why am I still seeing ads on mobile? ### [May 2, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1785843950995308553) > life could be so much worse. like i could be living in europe during the AI revolution. damn, I shudder at the nightmare ### [May 1, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1785674261522506222) > i'm growing the opinion of AI actually making more jobs than currently available; it will know you. know what you excel at. know what you love to do. it will put you in a profession that you can excel at. > > it will grow your love for whatever you have passion in, and give you a job there. ### [May 1, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1785659757153640487) > stronghold crusader was not just an amazing game, but an amazing feat of software engineering. it ran on (compared to today) potato computers without degradation to gaming experience. > > build software like that. ### [May 1, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1785475074520391990) > I’m getting really sick of seeing unsubstantiated engagement farming guesses on my timeline. > > “I think GPT2 is actually GPT-5\! Damn it’s not as impressive as I thought it would be…” > > You’re either an idiot, engagement farming, or both. > > I hope people start seeing through these charlatans more, but judging by the comments on these types of tweets… ### [Apr 30, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1785300602907656534) > for some reason black coffee makes me sleepy. only red bull / monster energy drinks give me a good enough kick. maybe i'm caffeine fatigued, if that's a thing... ### [Apr 30, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1785164984512901558) > the state of AI twitter [Attached image](/x/media/e20bae27d48e7864172269a05ba3ad7027ab6950dfe0cdde71342d5db421b427.png) Quoted post by Phil: ### [Apr 29, 2024 — @phill__1](https://x.com/phill__1/status/1784964135920235000) > Whatever gpt2-chatbot might be, it definitely feels like gpt4.5. It has insane domain knowledge I have never seen before ### [Apr 30, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1785127816549527619) > I have yet to see an impressive example from this gpt2-chatbot. cool, it makes ascii art that's part of its training set; but where are the mindblowing results? > > we can't really test out context length / retrieval on it yet; maybe that's where it's mindblowing. > > hopefully. ### [Apr 29, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1784899170538532911) > I saw nothing but Apple Vision Pro on m timeline for 5 days straight; after those 5 days, I have seen nothing about Apple Vision Pro > > Dead after 5 days ### [Apr 29, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1784834953718677816) > OG AI, iykyk [Attached image](/x/media/766937b8a02a2f701c89ffc6c3e5ef78c8603801fbc69e5d57abb74012936f47.png) ### [Apr 29, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1784832502932349420) > anyone who tells you there's no difference between a $10 mouse and a $100 mouse is a damn liar ### [Apr 29, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1784773722316161342) > guys guys guys, we'll be fine if there are no more human generations after this last one, who cares if all of humanity stops reproducing, everything will be ok > > LMAO what an L take Quoted post by Spencer Schiff: ### [Apr 29, 2024 — @SpencerKSchiff](https://x.com/SpencerKSchiff/status/1784769235861766278) > [@elonmusk]() This is flat-out wrong. Human civilization will be fine even if no more humans are ever born starting tomorrow. There’s nothing wrong with natalism but it’s weird to be obsessed with it and to say things that contradict other aspects of your worldview (like ASI being imminent) ### [Apr 29, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1784773073532793101) > this has got to be some meta ironic post from spencer. no way he actually thinks this way. > > props if you're joking - if you're not, holy fucking YIKES. [Attached image](/x/media/0b895884875253ef0f069c4b0dcfa47cdbdc87823759a4b3df96f658cec03b8c.png) ### [Apr 28, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1784470795118641316) > remember anon, you're only a few multiplications away from $10K per month ### [Apr 28, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1784379883667456076) > GPT-4-Turbo is as fast for me now as GPT-3.5-Turbo was. Something changed in OpenAI's inference speeds. ### [Apr 27, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1784193489015808003) > any app that requires an invite, certain-device-only, no web app, no way to use it via desktop, is DOA > > ESPECIALLY a social media app Quoted post by sankalp: ### [Apr 27, 2024 — @dejavucoder](https://x.com/dejavucoder/status/1784187554436378688) > every 2 days i post this on airchat [Attached image](/x/media/f7016f09e20a216bde464b61e67ab30487fdbda84fadfedb904be741554384a1.jpg) ### [Apr 27, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1784122353561252079) > everything is 20 bucks a month > > \>claude 3 opus $20 > \>chatgpt $20 > \>twitter blue $20 > \>google one $20 > \>apple one $20 > > that's $100/m, $1,200/yr btw > > not counting my calendly, zoom, skype, google meet, netflix, disney plus, youtube premium, digitalocean... > > altogether probably $300/m. ### [Apr 27, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1784120468221706347) > I wish I could just buy more tokens for [@AnthropicAI]() Claude Opus instead of having to make another account, membership again, etc... ### [Apr 27, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1784066566939939300) > remember; AI currently is really good at adding code, and terrible at refactoring/removing code. just keep that in mind when you're developing your codebase ### [Apr 26, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1783654409567633680) > agreed, this is a fun legal thought experiment. at what point can you still call it plagiarism/copyright theft? 10% from-humans data? what if it's 1% from-human data, 99% ai-generated data? what if it's 100% ai-generated data? Quoted post by Spencer Schiff: ### [Apr 26, 2024 — @SpencerKSchiff](https://x.com/SpencerKSchiff/status/1783651112341352500) > I don’t think human-generated data will continue to be valuable for training AI. Humans (even the smartest ones) have low intelligence and low creativity in the grand scheme of things. AI-generated data will soon be far more valuable. Which means the copyright debate is pointless ### [Apr 26, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1783653338224959596) > my Obsidian plugin SystemSculpt AI is now live on the Community Plugins place, go check it out ;) > [obsidian.md/plugins?id=sys…]() ### [Apr 25, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1783436052855705899) > i have not seen, to date, 1 person actually using an AI pin (humane) and say "damn that's cool" other than their own marketing team lmao ### [Apr 25, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1783399864778244260) > I really wanna try Open ELM from Apple. ### [Apr 25, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1783290871766724614) > don't worry anon, you'll be able to make an OF soon too and rake in the cash just like everyone else. equality 🤝 Quoted post by AI KATANA: ### [Apr 24, 2024 — @ai_katana](https://x.com/ai_katana/status/1782971948387217907) > I’m scared [Video on X](/x/media/0587ca682c63cea32997a2228d05d9dfe74e0792ca0afafccc3c41e08e8967c8.jpg) ### [Apr 24, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1783260083583148035) > google glass had a longer hype window than apple vision pro ### [Apr 24, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1782990974123376784) > yeah as usual you can see that phi 3 is very obviously trained on OpenAI synthetic data / responses, peep the OpenAI mention in the refusal [Attached image](/x/media/aed40b9ec0817fe16eef35a25b3519f783e9bccc11d2c2c714eb41854c24798b.jpg) ### [Apr 24, 2024 — @MichaelStolarz](https://x.com/MichaelStolarz/status/1782976872055513196) > this phi 3 mini model is kicking gpt 3.5 turbo's ass in some tests I've been giving it. insane to think a 3.8B model is this good, can't wait to test phi 3 small 7B. [Older posts](/x/before/1713929787000-1782976872055513196/index.md) [More from @michaelstolarz](https://x.com/michaelstolarz) Saved from X: 2026-09-07T04:06:23.190Z