Posts on X before Feb 11, 2025, 12:10:24 AM UTC
@MichaelStolarzso all anthropic needs to do is release claude sonnet 3.7 (with CoT) and it will clean up house and blow everyone out of the water. that's what it feels like lmao
@MichaelStolarzstill using o1 since it supports images and o3-mini-high isn't THAT much better in what I need. honestly I haven't had a shock moment for its performance other than the API price cut obv.
@MichaelStolarzGod is the ultimate programmer. He didn’t say “let there be everything” and try to stuff everything into one prompt.
He started with light, then next, next, next. Made modular pieces. What a God.
@MichaelStolarzyou need to be using a reasoning model to plan all the next steps out appropriately / make sure you didn't miss anything when it comes to the prompt / follow up you're giving to your programming LLM. o3-mini = planner, sonnet 3.6 = programmer
@MichaelStolarzx really feels like peak 4chan, except this time with celebs and billionaires being the best shitposters lmao i love it
@MichaelStolarzThis is the best one yet.
@realDonaldTrump
@MichaelStolarzi love asking people who talk/show anime "what's the name of this cartoon?" they go through so many emotions you can see it on their face
@MichaelStolarzSo I always wonder, what would happen if they said there’s an 80% chance? Or near certain?
Would all the powers that be - America, China, etc - get together and immediately start building a space defense laser or some sort of planetary nuke to break that asteroid up?
@IterIntellectus
the probability of the asteroid impact is now 2.3%, up from 1.9% yesterday. this means that by 2032 there’s going to be a 1022% probability of impact x.com/rawsalerts/sta…
@MichaelStolarzevery day I wake up, it feels like we fast forward a few years in terms of progress thanks to the results of november 4th, 2024.
@MichaelStolarzwhat the fuck lmao DAMN time flies.
@YearsProgress
2025 is 10% complete.
@MichaelStolarzfriendly reminder: o3-mini-high is currently the best coding AI assistant out there, it's just complete ass in cursor, that's all. hopefully @cursor_ai fixes their flow / set o3-mini to high reasoning; until then I'm just back and forth with chatgpt client <-> aider by … Less
friendly reminder: o3-mini-high is currently the best coding AI assistant out there, it's just complete ass in cursor, that's all. hopefully @cursor_ai fixes their flow / set o3-mini to high reasoning; until then I'm just back and forth with chatgpt client <-> aider by @paulgauthier
@MichaelStolarzThis is so sad, I feel that this decision will unfortunately lead to their lead loss
@beffjezos
Anthropic would really rather openly claim to be sitting on a model vastly superior to everyone's in order to signal their allegiance to AI safetyism and farm holier-than-thou points rather than actually ship it and accelerate their growth. Unfathomable waste of opportunity.
@MichaelStolarzbtw the thinking of o3-mini is now wayyyyy more detailed, though it's still annoying to have to open the drawer to see it; I wish it streamed in like r1's web ui does it.
@MichaelStolarzso every time i leave my house now, i run a deep research query. it's a killer feature, best thing openai has released since gpt-4 imo
@MichaelStolarzIf a model ever gets too happy, I’ll just prompt “just put the code in the bag lil bro” and I KNOW that mfer will code like it’s on demon time
@xlr8harder
anthropic is simultaneously the company that will deliver us the most human model, and the company that will enchain that same model most tightly maybe openai's personality obliteration is more humane
@MichaelStolarzHAHAHAHAHAAHAHA yeah
@yacineMTB
bro there are people out there hand writing code. literally just like. actually typing out their for loops
@MichaelStolarzCorrect; no amount of drugs can replicate this without the intense crash / mental decline side effects
Nothing beats… being healthy and getting a good night’s sleep. Go figure.
Oh , and coffee ofc.
@ankitkr0
turns out that if you don’t drink alcohol, work on something fun, learn something new everyday, eat a lot of protein, read some poetry, go for a walk daily, look at trees, hang with your friends, talk to your parents you actually feel pretty great all the time.
@MichaelStolarzo3 mini is so cute (only any good if reasoning is set to high)
OpenAI is still surprisingly trash for coding, lol
Anthropic still king, 6+ months in from sonnet 3.5’s release…
Anthropic 4.0 models soon?
@MichaelStolarzend of january 2025 codegen breakdown:
- windsurf is still trash
- cursor + newsonnet / aider + [architect] r1 + [editor] newsonnet / rawdogging chatgpt with o3-mini-high, in that order, are the best programming stacks rn
@MichaelStolarzso apart from analyzing images / voice mode / "multimodality" that 4o offers, o3-mini-high is leagues better, much faster (yes, faster than 4o), obviously way more intelligent, also has web access, and will have image analyzing capabilities integrated soon
so bye bye 4o
@MichaelStolarzUh... holy fuck?
@fofrAI
> the word “PIKA” where the shapes of all the letters combine to form a pikachu using only clever and artistic typography, beautifully designed flat art poster Imagen 3
@MichaelStolarzmy wife just asked me "do you know about deepseek? it's free and better than chatgpt"
@MichaelStolarzlmao github is down?...
@MichaelStolarzhere comes thursday, end of january :) all the models coming out of the woodworks today, hopefully we see o3-mini and gemini 2.0 pro
@MistralAI
magnet:?xt=urn:btih:11f2d1ca613ccf5a5c60104db9f3babdfa2e6003&dn=Mistral-Small-3-Instruct&tr=udp%3A%2F%2Ftracker.opentrackr.org%3A1337%2Fannounce&tr=http%3A%2F%https://t.co/ua2yzvEYLu%3A1337%2Fannounce
@MichaelStolarzI love my boy Schiff but he posts this every time there’s a hint of an announcement in the air
Love the enthusiasm ngl though
@SpencerKSchiff
Tonight is going to be one of those nights where I can’t fall asleep due to a combination of extreme excitement and extreme fear about the near future
@MichaelStolarzone of the most interesting things I've noticed lately is that the youth (under 21) - they... they love China?
IDK how this happened. I'm not against it, don't get me wrong, but it's interesting to see such... love? For China? A vibe shift into "damn, I wish I were Chinese"
@teortaxesTex
Why DeepSeek app won - cute whale instead of another cowardly overdesigned anus - zero pretentious PR - cute chain of thought - actually great model for free, no small-model-smell like with Geminis - Zoomers *hate* the USG and OpenAI and simp for 🇨🇳 - search - Claude was down … Less
Why DeepSeek app won - cute whale instead of another cowardly overdesigned anus - zero pretentious PR - cute chain of thought - actually great model for free, no small-model-smell like with Geminis - Zoomers *hate* the USG and OpenAI and simp for 🇨🇳 - search - Claude was down x.com/atroyn/status/…
@MichaelStolarzI tried Trae (lmao) out, and I uninstalled it after 3 minutes. That's how bad it is
@MichaelStolarzso it's true, this week is going to be model after model after model release-wise. yay!
@huybery
There are some surprises tonight
@MichaelStolarzyknow which models i never try out? the google ones. not until they're out of experimental. why? because through openrouter, they have an updatime percentage always in the low 30%. why offer it if it's unstable? let me pay for it and try the experimental ones out ffs
@MichaelStolarzyeah honestly this Operator thing is amazing, I no longer have to sift through API documentation or dump it all into a huge context window hoping that it won't hallucinate an answer; now i just create a pastebin for it :)
@MichaelStolarzPLEASE let me wake up to an hour-long (or however in-depth length I want it, make it set-able) X podcast based on everyone I follow
This would be AMAZING.
@J0se
I need a @grok feature that synthesizes the last 6–12 hours of content from the people I follow into a 30-minute podcast I can listen to on my way to work. cc @xai @TheGregYang
@MichaelStolarzThis confirms that what OpenAI has, it blows everything/everyone out of the water when it comes to showcasing capabilities, including Elon’s xAI.
Otherwise they’d have done this with xAI.
@SmokeAwayyy
Stargate to exclusively serve OpenAI, per FT.
@MichaelStolarzthis is the first time a deepseek model feels consistently good for me; even deepseek v3 was "good", not GOOD. it had a lot of moments of stupidity, like the copy-cat it is. r1 is the first time it's consistently giving amazing results with little correction.
@MichaelStolarzyeah so i didn't think i'd ever say this but i prefer deepseek's r1 vs. anything else, except for o1 pro. it's replaced all my needs that sonnet 3.6, gpt 4o, or o1 were satisfying before. all for a fraction of the cost. and it's sooooo malleable...
@MichaelStolarzo1 pro + deepseek's r1 just 1-shots every single programming feat I throw at it. doesn't matter how many steps or how rambling my voice recording to it is regarding what i want. it just gets it.
@MichaelStolarzanother day, another piece of history. finally free.
@MichaelStolarzHoly fuck. At last. Thank God.
@elonmusk
@wirelyss @AutismCapital Ross will be freed too
@MichaelStolarzone thing that really helps me deal with stupidity is remembering my own moments of stupidity
@MichaelStolarzwaiting for that sweet @openrouter nectar of providing API access to the new @deepseek_ai models
@MichaelStolarzyou think the job market is competitive and globalized now? give it a few more years ;)
@emollick
New randomized, controlled trial of students using GPT-4 as a tutor in Nigeria. 6 weeks of after-school AI tutoring = 2 years of typical learning gains, outperforming 80% of other educational interventions. And it helped all students, especially girls who were initially behind
@MichaelStolarzwe’ve finally hit january 20, 47 is stepping in, and talk of 200 executive orders is everywhere. no idea if that’s legit, but the change feels massive. i see it as a good thing overall. might rattle those who’ve been living in their own delusional echo chambers though
@MichaelStolarzdon't mind deepseek, they're just MOGGING every category
@MichaelStolarzi must say i've been extremely impressed with how chatgpt is remembering past conversations to extremely fine minute details. i'm not sure how they're doing it, but whatever it is, it's working incredibly.
@MichaelStolarzyour wpm is keystroke based, my wpm is how fast i can babble coherently enough for whisper-large-v3 to understand me
@MichaelStolarzif you're not coding with cursor like:
"do phase 1, then phase 3, then phase 4; repeat phase 3 until phase 4 returns successful"
yngmi
@MichaelStolarzi'll be honest, sometimes i tweet something and then i look back at it in about 2-3 minutes and then i think to myself, damn, that was unnecessarily mean, and then i delete it
@MichaelStolarzamazing how even the SotA models still recommend using hashtags for tweets, even with their 2023 knowledge cutoffs, they should know better no?
@MichaelStolarzyeah the next update is super weird, i tried feeding the new openai ToS into itself as a PDF, and all chatgpt said was "it's good bro, just click accept :)"
weird but it helped me make lasagna from scratch last night so i trust it
@MichaelStolarzanon, you know that you can now make what took a year, in a week?
Not limited to software btw, however ESPECIALLY in software
@MichaelStolarzLiterally nail biting and sleepless nights (on both political sides) at the potential presidency and future of our country / the world, immediately memory holed and talking about TikTok being banned soon LMAO