MS.

Posts on X before Dec 7, 2024, 9:11:12 AM UTC

On X

@michaelstolarz ↗
@MichaelStolarz
One thing I’ve come to realize (fortunately and unfortunately) is that we’re at a point where we can build anything software wise. And I mean that quite literally. If you know how to use agents, self improvement, and just replace your main driver model with whatever’s the latest … MoreLess
One thing I’ve come to realize (fortunately and unfortunately) is that we’re at a point where we can build anything software wise. And I mean that quite literally. If you know how to use agents, self improvement, and just replace your main driver model with whatever’s the latest SOTA… then yeah, there’s nothing you can’t do It’s still difficult, but for those that weren’t experts in certain domains, it would take years of study to apply things in a new domain. Now that same level of quality work can be done in a few months
@MichaelStolarz
Interesting to see that no one has reached limits yet with o1 pro… super interesting. Might just have to try it out
@MichaelStolarz
The real white pill is realizing Safari is the best browser if you’re deep into the Apple ecosystem already Few.
@MichaelStolarz
Hahahaha nope, THAT mode is for gooberment / internal use only
@MaxRovensky
@sama Does the $200 mode remove the thing where it lectures me on morality instead of answering a fucking question?
@MichaelStolarz
personally i don't care much about chatgpt itself, i care about what i'll have access to regarding the API fingers crossed that these new "o1-pro" and "gpt-4.5" will be available through API usage - idc at what price, I just want to test them in my own environment/sys prompts
@MichaelStolarz
it really feels like major softwares of our time with 100+ team members are actually just churning out unoptimized "uniquely designed" slop just so that they ship something reminds me of game studios
@chamath
Is it just me or was 18.1.1 a total turd upgrade. Phone doesn’t work anymore.
@MichaelStolarz
this +: - AR (ex. apple vision) - speech to action - 3d scene realtime rendering - suno-level audio generation -character consistency/“models” then all you need is a dedicated empty room to build in; walk around, design a movie, make a game, i mean you’re literally tony stark
@krea_ai
announcing Krea Editor. our new editing tool feels like magic. who wants beta access? 👇
Media attached to this post
@MichaelStolarz
i've been seeing a lot of rumblings from openai employees (i analyze their post frequency via their notis being turned on for me) it's matching the same frequencies as whenever there's a release coming, so i'm pretty sure something releases this week, maybe even today (tuesday)
Media attached to this post
@MichaelStolarz
TIL perplexity API is pretty useless because there's no Perplexity Pro equivalent (ex. reasoning, ability to use Sonnet 3.6 as the search model, etc) Just some outdated llama 3.1 perplexity online models. booooo 😭 @perplexity_ai @AravSrinivas pls implement, I'll pay w/e
@MichaelStolarz
bro, just one more all-nighter, bro. the api rewrite will be so worth it, bro. cleaner codebase, bro. just one more, bro. i swear, bro
@MichaelStolarz
surprisingly, the only 99.9% accurate speech-to-text, of course, after i do a post-transcription on it, is still OpenAI's Whisper API. all of these other providers that offer Turbo, like Groq, or any of the versions that are hostable on your own, if it's not that good ol' Whispe… MoreLess
surprisingly, the only 99.9% accurate speech-to-text, of course, after i do a post-transcription on it, is still OpenAI's Whisper API. all of these other providers that offer Turbo, like Groq, or any of the versions that are hostable on your own, if it's not that good ol' Whisper, it's just not as accurate. what i find that happens very often is, it'll just not transcribe a portion of your speech. i'm not sure why this happens. it'll just skip over things, or it'll do the notorious "..." instead of actually transcribing out the words. it's almost as if it's too lazy to do the full transcription sometimes. really interesting behavior. hopefully it's fixed in V4, or whatever. but for now, if i want to have a good focus flow, without having to go back and try and fix transcriptions and such, or even worse, having to re-say everything that i initially said, i'm just going to stick with OpenAI's Whisper API for now. and then of course, Sonnet 3.6 for post-transcription touch-ups.
@MichaelStolarz
when i first started using Cursor a little over a year ago, i really felt the "10x programmer" meme. especially since i was already so comfortable with the terminal and setting up automations in a pseudo-agent type of way with file watchers and procedures and loops... you get th… MoreLess
when i first started using Cursor a little over a year ago, i really felt the "10x programmer" meme. especially since i was already so comfortable with the terminal and setting up automations in a pseudo-agent type of way with file watchers and procedures and loops... you get the idea. now, with the latest Cursor update, where it has the agent feature, that initial 10x feels like another 10x. it's a steep learning curve, which isn't obvious from a first glance, even for those that use Cursor or code-gen tools every day. i think 99% of programmers don't utilize Cursor efficiently. and then 99% of the 1% that do utilize Cursor don't really think through how this new agent feature is as big of a game changer as code generation AI is. once people catch on and realize how powerful this is, i don't see any software limitations in terms of what can be created, even if tech progress magically stopped and we never have a better model than Sonnet 3.6 (lmao but you get the point) what an amazing time to be alive, waiting for the next state-of-the-art LLM to further this productivity explosion, and more importantly imo, even better workflows that no one has even thought of yet (but AI will for you, especially when personalization becomes an integral part, not even mentioning continuous learning...)
@MichaelStolarz
"Almost every important piece of software from the last 10 years uses Go"
GIF attached to this post
@wispem_wantex
Go is the programming language of the internet and the modern era. Almost every important piece of software from the last 10 years uses Go: - Docker - Kubernetes - Terraform - Prometheus - Grafana - Lets Encrypt - CockroachDB - etcd - Ethereum - Vault - Packer - Caddy
@MichaelStolarz
2024 isn't over yet, so i really hope i'm wrong with this next statement, but here it is: the only notable advancements that you can really feel in terms of output improvement have been Sonnet 3.5 and the o1/r1 "reasoning" models. everything else is cool, but tbh it just kind o… MoreLess
2024 isn't over yet, so i really hope i'm wrong with this next statement, but here it is: the only notable advancements that you can really feel in terms of output improvement have been Sonnet 3.5 and the o1/r1 "reasoning" models. everything else is cool, but tbh it just kind of seems slopped together
@MichaelStolarz
TLDR: pretty much every frontier AI lab has perfected AI text generation slop and so now we can finally focus on what really matters
@_jasonwei
Prediction: within the next year there will be a pretty sharp transition of focus in AI from general user adoption to the ability to accelerate science and engineering. For the past two years it has been about user base and general adoption across the public. This is very natura… MoreLess
Prediction: within the next year there will be a pretty sharp transition of focus in AI from general user adoption to the ability to accelerate science and engineering. For the past two years it has been about user base and general adoption across the public. This is very natural because user growth is a critical part any business model. But at this point I would say that there is widespread accessibility of LLMs: for most queries from the average person on earth, many LLMs can answer pretty well. In the upcoming five years I think the focal point will be the ability for AI to accelerate engineering and scientific research, which is the engine of progress in technology. At the frontier of innovation in any field, by definition there will be many open questions and a lot of headroom for better AI to make a difference. The stakes will be very high because progress compounds and also because AI accelerating AI research itself is a strong positive feedback loop. The other way of saying this is that there is somewhat limited headroom for improving the average user query, but massive headroom for improving the experience for the 1% of queries that would accelerate technological advancement, as well as on queries that people would want to ask the model but currently don’t because models are not smart enough to answer. AI research tends to improve where there is great headroom, and in scientific innovation there be substantial upside.
@MichaelStolarz
They're not wrong. It's just not the "open" that we're all used to. Open here means "open to vulnerabilities," not open source. Definitely not open source.
@tomwarren
quite the claim from Microsoft here 🤔
Media attached to this post
@MichaelStolarz
If you’re ever feeling down and want a stark reminder as to how far ahead you are of everyone else when it comes to AI-anything, just take a look at the replies in this thread
@_lizharvey
Maybe im stupid (im not) but there is literally not a single thing id need chatgpt for??????????? like does it do the dishes? x.com/ilikefufu_/sta…
@MichaelStolarz
It's so funny to see the annual cycle of people finding out about some new Twitter/X replacement... ...followed by them flocking over there, convincing their friends and family to use it... ...they come back to Twitter and post about it, saying "Oh my gosh, the discourse there … MoreLess
It's so funny to see the annual cycle of people finding out about some new Twitter/X replacement... ...followed by them flocking over there, convincing their friends and family to use it... ...they come back to Twitter and post about it, saying "Oh my gosh, the discourse there is so amazing!"... ...and then this lasts about two-three weeks, after which they silently come back to being full-time on Twitter once again. It takes them a while to realize how much they take for granted, and pair that with the fact that the honeymoon phase of using a new app / getting so many likes / comments / engagement with the thought of "wow, I just started, too, and I'm getting so much engagement already!"... All of it fades away and yeah, X will always take you back in. Welcome back.
@MichaelStolarz
It never ceases to amaze me how people seem to avoid the reality of impending changes, whatever they may be. It seems like, even against all odds, they still hold on to the belief that everything will remain the same. In pretty much every field I can think of, whether it's po… MoreLess
It never ceases to amaze me how people seem to avoid the reality of impending changes, whatever they may be. It seems like, even against all odds, they still hold on to the belief that everything will remain the same. In pretty much every field I can think of, whether it's politics, societal structure, software, biology - I change my mind frequently given new information. That's the whole point of progress, right? I consistently check to see if I have any static beliefs, and I make sure that I always have worthy reasons to believe what I believe. If not, I go down a rabbit hole of research, playing both the defense and prosecutor positions before I finalize my decision. I've trained myself to enjoy changing my mind on something that feels almost hard-coded into me - it honestly feels like a software update that fixed some bug in my code, to put it crudely. Most people, from what I can tell, are the opposite - they'll take their opinions and decisions, even if they know they're wrong, to the grave. They rather go never saying "you were right" - or worse, "I was wrong".
@MichaelStolarz
Big as always from DeepSeek.
@deepseek_ai
🚀 DeepSeek-R1-Lite-Preview is now live: unleashing supercharged reasoning power! 🔍 o1-preview-level performance on AIME & MATH benchmarks. 💡 Transparent thought process in real-time. 🛠️ Open-source models & API coming soon! 🌐 Try it now at chat.deepseek.com #DeepSeek
Media attached to this post
@MichaelStolarz
Anthropic's Haiku 3.5 would be justified with a 4x price increase if it supported images. At the very least, they should have kept the price the same as it was, stating that they're working on image capabilities and then increase the price once that goes into effect. All they'v… MoreLess
Anthropic's Haiku 3.5 would be justified with a 4x price increase if it supported images. At the very least, they should have kept the price the same as it was, stating that they're working on image capabilities and then increase the price once that goes into effect. All they've really done is shoot themselves in the foot by neither releasing image capabilities nor keeping the price the same, which has led to negative publicity and, for lack of a better term, a failed launch for their "mass use" model.
@MichaelStolarz
While OpenAI's ChatGPT search has been great for quick questions and straight-to-the-point answers, it fails miserably on any type of follow-up you may have. Unless your follow-up is a question about either something else entirely or for something that truly wouldn't make sense … MoreLess
While OpenAI's ChatGPT search has been great for quick questions and straight-to-the-point answers, it fails miserably on any type of follow-up you may have. Unless your follow-up is a question about either something else entirely or for something that truly wouldn't make sense for ChatGPT to repeat itself, it will always do exactly that—repeat itself. You can see in the screenshot here, all it really does is change a few words, and then it regurgitates the exact same thing, even though I specifically mentioned that it's missing something. I'd have thought that by now they would have fixed this, as ChatGPT has been around for, what, weeks? Has it been a month yet? Nevertheless, it should have been squashed and hot-fixed because surely this is annoying to more than just myself.
Media attached to this post
@MichaelStolarz
I think one of the most influential things that Elon's new DOGE project can do is introduce an evolution to the way that Americans pay their taxes every single year. I want to know how much I owe and pay it like a bill. I don't want to use paid services. I don't want to guess… MoreLess
I think one of the most influential things that Elon's new DOGE project can do is introduce an evolution to the way that Americans pay their taxes every single year. I want to know how much I owe and pay it like a bill. I don't want to use paid services. I don't want to guess. I don't want to worry about penalties down the road because something slips my mind. Who honestly believes that the IRS doesn't know how much you owe? I just want to have a hub where I can dump all of my receipts, expenses, and income streams, whether they come from client work or a 9-to-5 job or the gig economy or crypto investments or... you get the idea. It would remove literal months of stress, as well as planning and execution time, from everyone's life every single year. I mean, come on, we literally name it "tax season" because it takes a season to go through it and get it all done. It's insane to me, and remember, this is all being said without mentioning AI because we all know that it's going to be decades before everyday government services make any practical use of these emerging technologies. @elonmusk @DOGE
@MichaelStolarz
the funniest / most fun timeline would be if this tweet were true and not made in jest. imagine, riemann's hypothesis confirmed being solved through a tweet-er, post.
@hyhieu226
Grok-3 just proved Riemann's hypothesis. We decided to pause its training to check its proof, and if the proof is correct, training won't be resumed, as the AI is deemed so smart that it becomes a danger to humanity.
@MichaelStolarz
I've completely removed using GPT-4o from any of my workflows. I found that GPT-4o-mini works just as well. Rarely do I find room for improvement from whatever processing results I'm looking for. For anything I need to designate more brainpower to, I use O1-mini first, and … MoreLess
I've completely removed using GPT-4o from any of my workflows. I found that GPT-4o-mini works just as well. Rarely do I find room for improvement from whatever processing results I'm looking for. For anything I need to designate more brainpower to, I use O1-mini first, and in rare cases, I'll run it just to double-check through O1-preview. Absolutely everything else I use Claude Sonnet 3.5.
@MichaelStolarz
2025 is really shaping up to be the year of mainstream robotics. 2024 has been one very long teaser of what's to come. In 2025, we're probably going to see useful general-purpose robotics for under $5,000.
Media attached to this post
@MichaelStolarz
It's such a no-brainer that anytime I set up search for anything, I focus on making sure all setting names, as well as short descriptions (if available), are able to be fuzzy found. Another fallback for me is usually ripgrep. Search results should almost never be empty.
@d4m1n
I feel like no matter how much AI we'll have we won't ever solve Search, Printers and LAN setup
Media attached to this post
@MichaelStolarz
It's so interesting to see everyone talking about scaling thanks to Ilya's quote, but they don't seem to understand that there's different types of scaling. It just goes to show that a lot of people seem to rewrite high performing tweets in order to just boost for engagement rat… MoreLess
It's so interesting to see everyone talking about scaling thanks to Ilya's quote, but they don't seem to understand that there's different types of scaling. It just goes to show that a lot of people seem to rewrite high performing tweets in order to just boost for engagement rather than have something of substance.
@MichaelStolarz
It's so insane to me that so many people are reading Ilya's quote and the only thing that they're registering are the first and last parts. They completely remove the middle. All they see is "results from scaling up... have plateaued." Weird how no one is mentioning this is spe… MoreLess
It's so insane to me that so many people are reading Ilya's quote and the only thing that they're registering are the first and last parts. They completely remove the middle. All they see is "results from scaling up... have plateaued." Weird how no one is mentioning this is specifically only regarding pre-training. And more importantly, it only mentions ONE of the phases of training an AI model that uses a vast amount of UNLABELED data.
@ylecun
I don't wanna say "I told you so", but I told you so. Quote: "Ilya Sutskever, co-founder of AI labs Safe Superintelligence (SSI) and OpenAI, told Reuters recently that results from scaling up pre-training - the phase of training an AI model that uses a vast amount of unlabeled d… MoreLess
I don't wanna say "I told you so", but I told you so. Quote: "Ilya Sutskever, co-founder of AI labs Safe Superintelligence (SSI) and OpenAI, told Reuters recently that results from scaling up pre-training - the phase of training an AI model that uses a vast amount of unlabeled data to understand language patterns and structures - have plateaued." ... threads.net/@yannlecun/pos…
@MichaelStolarz
Where's the American equivalent of these? I don't think even Boston Dynamics has this kind of maneuverability. It seems so flawless. It edges on... Making me feel as if... These machines are alive, the way they self-correct. Balance. Plan and prepare their next movements. A… MoreLess
Where's the American equivalent of these? I don't think even Boston Dynamics has this kind of maneuverability. It seems so flawless. It edges on... Making me feel as if... These machines are alive, the way they self-correct. Balance. Plan and prepare their next movements. Again, I don't see this in any... mainstream American robotics company right now. Maybe we have all of this behind the scenes, but... it's just very weird to see this coming from random Chinese startups. If we have these behind the scenes, what does China have behind the scenes?
@burkov
You are not gonna hide from an army of these pic.x.com/dtFsvmr31i
Media attached to this post
@MichaelStolarz
probably a SOTA coder if none of the current SOTA labs release something new then again, it’s about that time for OpenAI’s early Christmas present… 5o? low hopes for anthropic since watching dario’s interview, especially knowing that we won’t be seeing opus 3.5 anytime soon
@JustinLin610
What will come next huh? x.com/Alibaba_Qwen/s…
@MichaelStolarz
if a government doesn't prioritize intelligence and its own hardware, it'll end up beholden to others. it's already happening, and the difference between prosperity and reliance will only increase (probably exponentially if we're looking at intelligence growth as the main factor)
@ns123abc
🚨🚨🚨Japan going full BASED: - $65B pumped into chips and AI -Rapidus is aiming at 2nm chips, gunning for 2nm tech to rival TSMC
Media attached to this post
@MichaelStolarz
i think we'll start seeing AAA studios buying out indie games just like monopolies buy out startups. Unfortunately for them though, I think they'll "polish up" the game after takeover, ironically making it worse after going through the AAA studio "upgrade / expand" filters.
@notch
Unless you count Bethesda as AAA, I don't remember the last time I played a AAA game. And shattered space made we want to quit bethesda too. Thankfully, the video game industry is doing just fine. Just play indie games.
@MichaelStolarz
i downloaded zen browser, used it for 3 minutes, and closed out of it. arc is still king for my use cases. maybe they will open source it? or their new v2 paid version of it will come out soon? idc if i have to pay $10 or whatever. it's a daily driver, why wouldn't I?
@MichaelStolarz
it's insane to me that o1-preview is 5x more expensive for input tokens and 4x more for output tokens compared to claude 3.6 sonnet... even though sonnet 3.6 is better (at least for coding/creativity).
@MichaelStolarz
seriously considering removing all of my AI service provider options from SystemSculpt (currently: OpenAI, Groq, Anthropic, Grok, OpenRouter, Local)... and just making it super simple: OpenRouter and Local. it's becoming more mainstream, has all of the latest models from all of… MoreLess
seriously considering removing all of my AI service provider options from SystemSculpt (currently: OpenAI, Groq, Anthropic, Grok, OpenRouter, Local)... and just making it super simple: OpenRouter and Local. it's becoming more mainstream, has all of the latest models from all of the top providers, a bunch of free models, little to no rate limits vs. native model use...
@MichaelStolarz
hopefully they fix this repetitiveness that often (very often) occurs specifically with SearchGPT. hey @OpenAI @sama I love it otherwise; it's my daily search driver since day 1 and I'm not looking back, only forward to even better improvements/evolution.
Media attached to this post
@MichaelStolarz
i used to drink nearly 2g of caffeine a day - no, not mg, not 2g of coffee, 2g of actual, pure caffeine - mostly in the form of white monster energy drinks (140mg caffeine each) and 30-40 grams of nescafe gold instant coffee (~40mg of caffeine in each gram)
@MichaelStolarz
using o1-preview, there's a "new version of chatgpt", giving me two reasoning replies. i think full o1 is coming super soon, maybe next week (or maybe today, thursday?)
Media attached to this post
@MichaelStolarz
after trying out @aide_dev by @skcd42 for a 1-file codebase for simple edits, trying out many different strategies of communication to see if my prompting may be the bottleneck, i can confidently say that it's nowhere near cursor-level quality nor efficacy. i do have high hopes … MoreLess
after trying out @aide_dev by @skcd42 for a 1-file codebase for simple edits, trying out many different strategies of communication to see if my prompting may be the bottleneck, i can confidently say that it's nowhere near cursor-level quality nor efficacy. i do have high hopes for it, or rather, its plan -> execute strategy shown in the sidebar. i will revisit it in 1 month.
Media attached to this post