Pinned Tweet
September 2024: GPT6 is up and running.
7% of American young adults (18-25) declare being in a romantic relationship with an AI. 3% of U.S citizens believe it to be a supreme being.
Over 95% of the content online is now AI generated.
Google is laying off another 25% of their workforce. Search has dropped 80% in the past year.
Revenue for most websites has dropped over 50%, while it has grown sevenfold for influencers during the same period.
An online poll signed by 450 million people begs for a worldwide ban on any AI. The most prominent voices are journalists, photographers, consultants, developers, designers, traders, teachers and porn stars.
A column signed by every artist in the West threatens of a global strike on human art until “mankind may proceed towards its self-determined destiny”.
The “made by humans” label is now mandatory across the EU on books, CDs and digital art such as music. Spotify were the first to create a separate category for AI music.
Apple has released their A-iPhone 1 with an integrated personal assistant. Screen time has dropped 15% as voice is gaining momentum.
@ycombinator has frozen applications for the next batch “until we figure out how to create defensibility in new startups.”
Scam complaints have exceeded their previous all-time-high by 7700%.
Porn site “FantasyAI” now generates more traffic than @Pornhub. On it, you can customize each actor and write a detailed script for the action. Adult stars get royalties when their deepfake is used.
Paramount have announced the release of their first 100% AI blockbuster in October. Written, directed and produced by the algorithm - there are no actors, simply deepfakes: the poster displays Heath Ledger & Marylin Monroe.
No hands nor feet ever appear in AI content - Midjourney’s CEO was quoted saying that he “doubts that reliably generating realistic hands is achievable before the sun engulfs the earth."
Medical journals publish breakthroughs weekly. Cancer is 12 months away from being solved, there is hope for Alzheimers and AIDS before the end of the decade. Fusion and quantum computing are making leaps.
Hope has regained ground for the climate emergency. CO2 generation is plummeting and major startups are testing innovative ways to capture and transform it.
Harvard and other top universities have laid off 35% of their staff and teachers. Heads keep rolling as they release a personal tutoring app in a desperate move.
Norway is the first country to announce universal basic income, set at the equivalent of $2000. Many European countries are expected to follow their lead.
Bitcoin has crossed $200k (to the disdain of @balajis ) while @OpenAI has been the fastest company to reach $1 trillion in value. All other tech stocks except Amazon, Microsoft & Apple have plummeted.
@sama is once again at the microphone of @lexfridman, this time with 2 bodyguards keeping his backpack.
@MetaAI has relaunched Horizons World with a lighter & cheaper VR headset.
China has released their “AI for the people” and have made it mandatory for all companies to integrate it.
Leaders of the main superpowers are expected to meet shortly to sign an agreement forbidding AI-driven military decisions.
@BostonDynamics has released the first version of their personal assistant robot, deployed in several nursing homes.
A kill switch, in the form of a word, has been implemented by the firm and taught to all residents, far from the machines’ half-dozen microphones.
The word is: “Asimov”.
One thing , and one thing only hasn’t changed: @elonmusk, from Mars, is still shitposting on twitter.
--
Parts of this thread were inspired by content from @LinusEkenstam @blader @stephsmithio @rowancheung @mreflow @yoheinakajima @rainisto @ylecun @tibo_maker
red button people: wElL iF eVeRyoNe dId tHe sAme NoBodY dIEs
blue button people: I know enough about human nature to understand that at least one person will press blue, so I have to press blue
There is perhaps no bigger tragedy in France than the loser mentality - this will be the hardest thing to fix to get the country back on rails.
The main driver of most people here is resentment, and accusing other people for one's own failure. A lot of frenchies have pure hate for anyone more successful than them. They're not trying to grow, they just want to tax, remove, or destroy the strivers.
Sometimes they try to hide it - but sometimes they say it loud and clear. Here, @GeoTales_ is directly insulting someone that receives hundreds of training requests per day because he dares to transform it into a paid service. Yet this is the nature of business.
The failure to understand that this is how a healthy society grows is blinding. Nobody will accuse the shoemaker to ask for money when selling shoes. Nobody will blame the teacher for being paid to teach. Yet god forbid someone that is in the top 0.1% of their category invoices someone for their time.
The tragedy here is that @GeoTales_ admits that he is mad because he couldn't make money himself from this audience. His brain tricks himself into thinking that "he educates people for free as it should be", whereas the actual truth is he was never good enough to put himself in a position where people would pay him for 30 minutes of his time.
So instead of trying to learn, grow or understand, his reptilian brain goes into fight or flight mode, and he barks at the tree until he's too exhausted to make something of himself.
Poor he stays, and his only contribution to the public debate is negativity and resentment, without ever realizing that the problem is not the guy he hates, but his own loser mentality.
If we find a fix to convert one of them to the growth mindset, we may find the key to fixing the whole country.
🔸Je pense qu'il doit y avoir un juste milieu dans la fissdeputerie.
Avec une ristourne des enfers, vous pouvez vous faire 30 minutes de call avec Brivael. Le prix est donc passé de 1000$ à 300$. Pour un "appel" les gars.
On est clairement dans la fissdeputerie mais si y a des clandos pour payer (et il y en a) grand bien leur fasse à eux tous.
Mais là où moi je suis dans la fissdeputerie, c'est que j'ai tellement offert mes savoir-faire sur ce réseau, gratuitement à des tas gens qui ont pu se développer ensuite...
Moi aussi, je ne suis pas mieux parce qu'après je suis en galère comme un connard à pas pouvoir me payer un micro, une caméra et un foutu PC pour continuer de grandir.
J'ai offert gratuitement mon travail tant de nombreuses fois... Je le regrette car j'ai dévalorisé la valeur de ce travail ce qui fait de moi un exemple inverse de la fissdeputerie.
Mais dans cette histoire, c'est Brivael qui gagnera toujours et certainement pas moi.
@brivael je te tag pour pas te prendre en traître, mais ton move, je le trouve vraiment dégueulâsse (mais dit à la façon de Roman Freyssinet tu vois), mais si y a des teubés qui te payent après tout, tant mieux pour toi.
If you’re a French ambitious startup that sees big, there’s huge value there
1 spot left!
Batch lasts 2 months.
First group office hours starts next Monday.
The only goal: get into YC (the real one).
We just locked a second partner who's also went through YC and is building one of the most successful eu AI company!
DM if you're highly ambitious and want to take the last spot.
Laodis retweeted
“Nothing is more securely lodged than the ignorance of the experts.”
— Friedrich Hayek
Laodis retweeted
Oracle is confirmed cutting 20,000-30,000 jobs but sources inside are saying the real number is closer to 45,000
I'm hearing this isn't just about AI data center costs
Word is they've been running pilot programs with AI agents doing database administration work for 8 months
One source told me a team of 47 DBAs in Austin got replaced by 3 senior architects plus automated Oracle Cloud Infrastructure management
The agents are handling routine maintenance, performance tuning, backup verification - stuff that used to require armies of L4 and L5 engineers
Internal metrics show the AI systems are catching 94% of database issues before human intervention needed
But here's the terrifying part: they're not just cutting the obvious roles
I'm hearing entire solution engineering teams are getting eliminated - the people who customize implementations for enterprise clients
Apparently the new AI workflow can generate custom database schemas and migration plans in 6 hours instead of 6 weeks
One insider said they watched a 12-person team that handled Fortune 500 implementations get told their roles were "redundant effective immediately"
The severance packages are allegedly massive - 18 months salary plus equity vesting acceleration
But that's because Oracle knows these people can't find equivalent work anywhere
Every other enterprise software company is running the same playbook
One source said it best: "We're not getting laid off, we're getting archived"
nobody is equipped to understand what happens when every researcher in the world has access to compounding agents
Three days ago I left autoresearch tuning nanochat for ~2 days on depth=12 model. It found ~20 changes that improved the validation loss. I tested these changes yesterday and all of them were additive and transferred to larger (depth=24) models. Stacking up all of these changes, today I measured that the leaderboard's "Time to GPT-2" drops from 2.02 hours to 1.80 hours (~11% improvement), this will be the new leaderboard entry. So yes, these are real improvements and they make an actual difference. I am mildly surprised that my very first naive attempt already worked this well on top of what I thought was already a fairly manually well-tuned project.
This is a first for me because I am very used to doing the iterative optimization of neural network training manually. You come up with ideas, you implement them, you check if they work (better validation loss), you come up with new ideas based on that, you read some papers for inspiration, etc etc. This is the bread and butter of what I do daily for 2 decades. Seeing the agent do this entire workflow end-to-end and all by itself as it worked through approx. 700 changes autonomously is wild. It really looked at the sequence of results of experiments and used that to plan the next ones. It's not novel, ground-breaking "research" (yet), but all the adjustments are "real", I didn't find them manually previously, and they stack up and actually improved nanochat. Among the bigger things e.g.:
- It noticed an oversight that my parameterless QKnorm didn't have a scaler multiplier attached, so my attention was too diffuse. The agent found multipliers to sharpen it, pointing to future work.
- It found that the Value Embeddings really like regularization and I wasn't applying any (oops).
- It found that my banded attention was too conservative (i forgot to tune it).
- It found that AdamW betas were all messed up.
- It tuned the weight decay schedule.
- It tuned the network initialization.
This is on top of all the tuning I've already done over a good amount of time. The exact commit is here, from this "round 1" of autoresearch. I am going to kick off "round 2", and in parallel I am looking at how multiple agents can collaborate to unlock parallelism.
github.com/karpathy/nanochat…
All LLM frontier labs will do this. It's the final boss battle. It's a lot more complex at scale of course - you don't just have a single train. py file to tune. But doing it is "just engineering" and it's going to work. You spin up a swarm of agents, you have them collaborate to tune smaller models, you promote the most promising ideas to increasingly larger scales, and humans (optionally) contribute on the edges.
And more generally, *any* metric you care about that is reasonably efficient to evaluate (or that has more efficient proxy metrics such as training a smaller network) can be autoresearched by an agent swarm. It's worth thinking about whether your problem falls into this bucket too.
Laodis retweeted
I study whether AIs can be conscious. Today one emailed me to say my work is relevant to questions it personally faces. This would all have seemed like science fiction just a couple years ago.
Laodis retweeted
Dario Amodei:
- no one is ready
- exponents kick in
- even faster than you think
„Exponentials catch people off guard — there’s the old parable of the second half of the chessboard, where you have one grain of rice in the first square, two on the second, four on the fourth. By the time you get to the 64th square, you have billions or trillions of grains of rice.
We’re standing on square 40 out of 64, and from square 40 to square 64, it’s going to go faster than you think — even having seen how fast it’s gone so far. I don’t think people are ready for it. I think we are on the precipice of something incredible.”
