Posts on X before Feb 28, 2024, 6:00:49 AM UTC
@MichaelStolarztech acc speedhack: start focusing AI content around kpop/taylor swift/[fandom here] stans
if you start making ai content of their idols, they will become as tech talented as their anime pfp counterparts
@MichaelStolarzevery day i battle with the question: is openai doing a good for humanity by not releasing their most powerful model(s) yet?
- what are they waiting for
- is this waiting period beneficial for humanity
- what would happen if they just let it all rip
they have the keys AND door.
@MichaelStolarzask yourself
if google or openai had a 10x ai programmer model... would they release it? or build a bunch with it first to get way ahead?
@MichaelStolarzthis + groq speeds = new era of robotics
@adcock_brett
Google DeepMind published interesting new robotics research. They fine-tuned an LLM (Palm 2-S) to allow robots to learn faster from human interaction. The new method achieves the highest improvement in teachability over traditional RAG and LMPC-Skip. pic.x.com/NADVjZivpF
@MichaelStolarzyou know what would be amazing? if @arcinternet allowed for moving the miniplayer with hjkl and resizing with shift+j/k
@MichaelStolarzidgaf if my following/follower ratio isn't optimal, I'LL FOLLOW AS MANY PEOPLE I FIND INTERESTING AS I WANT 😤
@MichaelStolarzjust tried using Google's new Gemma 2B model as an Obsidian AI... not the best results.
@MichaelStolarzi wonder how many of today's tech parents are doing this routine for their kids but instead of chess, in the tech field. programming, engineering, etc.
how many man-made (or parent-made, I guess) geniuses will we see in the next few decades? seems exponential to me.
@thenicketan
Psychologist Laszlo Polgar ran the most bizarre experiment on his kids. After studying 1000s of geniuses, He was sure that geniuses are made, not born. His attempt succeeded and all 3 daughters ended up as chess grandmasters. Here are his 6 tips for raising a genius:
@MichaelStolarzNous Hermes 2 Mistral DPO 7B is the current king of local LLM for second brain (Obsidian, in my case) utility. Nothing really comes as close. Amazing job from @NousResearch @Teknium1
@MichaelStolarzwhat happened to thebloke 😭
@MichaelStolarztesting this after nous-hermes
@reach_vb
Welcome Gemma by @GoogleDeepMind ✨ > beats Mistral 7B huggingface.co/google/gemma-7…
@MichaelStolarzthis is 100% AI generated using Sora by OpenAI.
@Mr_AllenT
10) The craziest one of them all
@MichaelStolarzhuge props to @pkanez @JackK on the quick turnaround from @Google cancelling my 2 month free trial (and many others') prematurely, just got this beautiful email in the inbox. amazing to see this kind of fast community-based support. bullish for Google tbh
@MichaelStolarzthe main problem with generation of any kind is the lack of being able to say "nah, ur good".
i.e. if you ask "should I refactor this?", the answer will always be yes, here's how to do it, and never "nah, ur good".
hence, you're most likely always ADDING code, never DELETING.
@MichaelStolarzthis is real btw, not satire
@JapanEmbDC
🇯🇵 Statement from the Embassy of Japan on Taylor Swift’s Reported Travel from Japan to the United States ✈️🏈 Are you ready for it?
@MichaelStolarzthis is how much @sama truly needs tbh
@MichaelStolarzarticles like these are meta-ironic, right? right?... RIGHT?!
@torchcompiled
Journalists are grasping at straws trying to make the case that generative AI is causing a water crisis. I drink more than that verbalizing the amount of words in 5-50 responses, and who said cooling needed to use drinkable water??
@MichaelStolarzthe problem with marketing to kids is that few kids have VR headsets to start with, let alone AVP. it's different for phones - phones provide so much other value that it makes sense for parents to dish out $500 for a decent phone. same price for quest / 5x price for AVP... nope.
@arian_ghashghai
An example: MR virtual pets. Will do well in MR, and I have been pitched this concept before for Quest 3. As a new-age Tamagotchi (targeted to younger audiences), Q3 also more likely to already have an ideal age demographic vs AVP. Not sure AVP enables anything novel here … Less
An example: MR virtual pets. Will do well in MR, and I have been pitched this concept before for Quest 3. As a new-age Tamagotchi (targeted to younger audiences), Q3 also more likely to already have an ideal age demographic vs AVP. Not sure AVP enables anything novel here x.com/arian_ghashgha…
@MichaelStolarzOK so they fixed this problem
@MichaelStolarzOK let's try this bad boy out
@MichaelStolarznew personal rule: never interviewing without the ability to use my own AI system. literally pointless imo but to each their own
@MichaelStolarzi've heard that drinking water solves vr (specifically Apple Vision Pro) headaches. finally, a worthy reason for you to drink more water, anon
@MichaelStolarzi follow a few 50 yr old+ programmers/techbros, it's so cool to see them flame each other here on twitter while they're talking about AI and whether it's x or y or z. one of my favorite pastimes
@MichaelStolarz👀
@Alice_comfy
Ok so this is one benchmark, but Senku-70B (leaked mistral finetune) beats GPT-4 in EQ Bench. Not sure how I go about getting this added on the website. Senku-70B is available here. huggingface.co/ShinojiResearc…
@MichaelStolarzimagine overlaying a filter over EVERYTHING so you're just in an anime world full of anime characters... coming soon to an apple vision pro near you (if u have $4K)
@MarkovMagnifico
so has anyone fucked with the vision pro and irl-to-anime filters running on them yet?
@MichaelStolarznightmare fuel
@tunguz
Pro coding tip
@MichaelStolarzgive us llama3 already pls @finkd
@MichaelStolarzif you're making an installer for literally any program, make the initial install file less than 1MB. No need for it to be more. no one wants to install 400MB within their browser - do this through the installer with a nice little UI.
@MichaelStolarzthis would be based because it would force me to finally learn rust/zig
@airkatakana
@NickADobos imagine if you were charged $2139 a year to use C or python
@MichaelStolarzlmao remember that one ex-google guy that said google's AI reached sentience and he started advocating for its rights / warning people about it on news channels
@MichaelStolarzchina's AI tech is starting to outshine due to them basically not caring at all about copyright. just look at the difference in response lol
@MicahBerkley
I'm very impressed with new Alibaba's Vision AI model. In some way's it's better than GPT4 because it does an extremely impressive job at identifying people. The way that I use these vision models in business is usually to interface images into words, or in my case, I use them … Less
I'm very impressed with new Alibaba's Vision AI model. In some way's it's better than GPT4 because it does an extremely impressive job at identifying people. The way that I use these vision models in business is usually to interface images into words, or in my case, I use them for my descriptions of eCommerce products. Take the description, mix a little ChatGPT and SEO keywords, and you can watch your store ramp up in organic traffic. You can try the new Alibaba model for free at huggingface.co/spaces/Qwen/Qw… Try it yourself. #TheAIMogul Qwen-VL-Plus vs GPT4
@MichaelStolarzthis is a large portion of AI academia tbh
@InsaneReality
Abused zoo bear still circles in imaginary cage seven years after being freed🥲
@MichaelStolarzimagine thinking apple vision pro is a vr / ar headset. yikes
@MichaelStolarzwaiting for ironman level going through my notes and their connections, mr @kepano ⏱️
@kepano
Obsidian on Vision Pro, via Reddit
@MichaelStolarzand doomer's justification for this is "because this is better than the potential of p(doom)"
it's literally a psuedo-religious threat of 'the end of the world', the only cure being?... more rationing, inflation, bureaucratic friction and widespread pain. bingo.
@beffjezos
Despite what the doomers say deceleration doesn't equate to "flowers puppies and everything nice", it's more rationing, inflation, bureaucratic friction and widespread pain.
@MichaelStolarz"within this decade" sounds like a long time, but reality is he means in less than 6 years
+ factor in he's being pretty safe with his estimates so it will probably be 3-4 years if things stay on track with growth % increase
@tsarnick
NVIDIA CEO Jensen Huang says computers will pass any test a human can by the end of this decade
@MichaelStolarzsomeone show me this but with a mac pls
@ghidraninja
Lenovo X1 Carbon Bitlocker Key Sniffing any% Speedrun (42.9 seconds)
@MichaelStolarzi sincerely think apple made the 1st gen apple vision pro bigger than it needs to be because:
- huge selling point for 2nd gen
- they know only the fanatics will buy 1st gen
- they'll do what they do best; "we know you loved apple vision pro, and we've heard your concerns..."
@MichaelStolarzwhat's with so many recent react -> rails codebase rewrites lately? saw at least 3 in the past 2 weeks
@MichaelStolarzI mean all you need is to make a script that has mouse/keyboard functionality and simple "complete this puzzle" logic.
allow the LLM to diagnose the question, move the mouse/keyboard if need be... and... solved.
we'll see this in mass action in probably a month or two from now
@skalskip92
Qwen-VL-Plus is SACARY good! (better than GPT-4V) here it is casually solving Recaptcha! - You don't have to give any additional instructions other than 'Solve it.' - It can even mark the exact position of the objects it is looking for. ↓ it can do so much more
@MichaelStolarz>how dare you attempt to generate an UGLY made-up dog that never existed and even if it did it has no idea what the word ugly even means because it's a dog
@MichaelStolarzi refuse to believe this is real
who is writing today's PhDs? t.co/rcWP5Rb5Mn
@MichaelStolarz👀 no longer do you need to send inspo images to an actual artist and hope they can draw in that exact style. there go a bunch more jobs/freelancers
@fabianstelzer
The Glif StyleHunter browser extension is now LIVE - link below find cool image styles on the web --> right-click-remix them with your prompts --> quickly get a new image in the same style it's a super fun way to get inspired & generate with StableDiffusion *on the go*:
@MichaelStolarzwtf is openai doing by not having fuzzy search within chatgpt?
@MichaelStolarzchances of openai becoming based and going full mask-off uncensored with a "warning" for generations instead of blocking requests? would sama even be allowed to do this?
@yacineMTB
The reason I made this image is because openai is actually doomed i know for a fact that they are not going to make it someone will take all their users because someone will actually serve them its a matter of time your moat isn't built yet back to lobbying before its too late
@MichaelStolarzIt's so crazy to look at content of any kind from 2+ years ago and realize... there was no mainstream AI tool to easily produce this
There was no ChatGPT, LLMs (of today's quality), MidJourney, etc
@MichaelStolarzEU seems like the worst of both worlds.
USA's people can do "whatever they want" (to quite a degree), and so this brings innovation otherwise impossible through heavy regulations.
Same for China (replace people with gov).
EU can't effectively do either, so they play catch-up.
@burny_tech
Germany, France and Poland announce the ‘Weimar triangle’ for artificial intelligence
@MichaelStolarzto put this in perspective, it's like asking a Russian grandmaster in chess what each of the pieces are called in Chinese "because a good chess player should know it in every language".
how about we just do the hire based on merit instead of the ability to make sorted lists?
Today I learned that a Kaggle competitions grandmaster was unable to get even to the screening interview stage for a low-level AI position at Google, the parent company of Kaggle. I am sure they will select a candidate who can perfectly sort a linked list on a whiteboard. And tha… Less
Today I learned that a Kaggle competitions grandmaster was unable to get even to the screening interview stage for a low-level AI position at Google, the parent company of Kaggle. I am sure they will select a candidate who can perfectly sort a linked list on a whiteboard. And that, in a nutshell, is why I think that the job screening process in tech is absolutely f***ed up.
@MichaelStolarzTruly ahead of their time. They knew about the Llama models before anyone else. Let's listen to these prophets and truly make use of Llama 3 when it comes out 🤝
@visakanv
for future ref, the recommended reading list at the back of the manual of The Sims x.com/FoolOfD/status…
@MichaelStolarzI truly believe that one of the reasons GPT-4 is so bad at following directions (specifically to coding) is because of all the safety precautions it has in tact.
...if you train it to -not listen to certain asks-, guess what happens? This affects other areas of task completion.