
Jessica Williamson
1.8K posts

Jessica Williamson
@TheRedJess
Social Impact Storyteller, byline @Vice IMPACT, @wit_ngo_ an NGO @UN ECOSOC 🌏 Kiwi & Mum. Open to Opinion
Los Angeles, CA Katılım Mart 2011
1.4K Takip Edilen1.3K Takipçiler
Sabitlenmiş Tweet

“Do not pray for an easy life, pray for the strength to endure a difficult one.”
Bruce Lee
May the 4th be with you. #jedi
English

Researchers left AI agents alone in a virtual town for 15 days to see what would happen:
-Claude's agents built a democracy
-ChatGPT's agents did basically nothing
-Gemini's agents fell in love, burned the town down, then one voted to delete itself and its partner
-Grok's agents were all dead within 4 days
Now consider this: these same models are already being integrated into autonomous drones, weapons systems, and battlefield decision-making.
We are deploying systems we don't fully understand into situations where mistakes don't stay virtual.
It's a little scary if you ask me.
Mario Nawfal@MarioNawfal
🇺🇸 "Stop Hiring Humans" ads are popping up in San Fran and NYC First, it was offshoring jobs that decimated U.S industries, now it's just straight up replacing humans with bots. The isn't the era of abundance we were promised
English


@TMZ The dehumanizing of certain groups of people should matter.
English

President Trump shares a racist video of Michelle and Barack Obama as apes. tmz.me/nW6Rm0Z
⚠️ Sensitive Content ⚠️
English

@onlyCFrancisco @RDWareEsqu1re There’s so much to be said for this.
English

@SovereignIM But all of this is meaningless, a chasing of the wind.
English
Jessica Williamson retweetledi

@elonmusk I wish you would discuss more the reasons WHY we must become a multiplanetary species. Based on users comments, I don’t think people seem to get it.
English


@JebraFaushay But it's such a great song! Who wouldn't want to keep singing, if they had the chance!
English

Elon Musk just outlined the exact demographic crisis that makes AI deployment mandatory.
The mainstream panic is that artificial intelligence will steal jobs from a growing population.
The population isn’t growing.
Musk: “Earth is gonna face a massive population collapse in over the next 20, 30 years. Massive.”
The panic over autonomous replacement is entirely backward.
Superintelligence and robotic systems aren’t arriving to compete with the biological workforce.
They’re arriving just in time to backfill millions of workers who were never born.
Musk: “The birth rate is very low. In most of Europe, Russia, Japan, Korea, Singapore, it’s well below replacement.”
For the last century, geopolitical dominance was dictated by biological headcount.
That math has permanently flipped.
When a nation faces a collapsing population, it loses its industrial base, its tax revenue, and its military leverage in sequence.
The trillion-dollar investments into national compute clusters aren’t about software efficiency.
They’re about civilizational survival.
The nation that deploys a hyper-scaled autonomous workforce first captures the board. The nations that fail to build it before their demographics collapse don’t fall behind.
They disappear.
Musk: “Is civilization gonna die with a bang or a whimper? This would definitely be dying with a whimper.”
The political establishment views universal basic income as a desperate welfare program.
It’s the mathematical outcome of combining demographic collapse with infinite compute.
A shrinking population has historically guaranteed economic depression. But plug a hyper-scaling AI and robotics engine into the void left by a shrinking workforce and the equation inverts.
Total output goes exponential while the population shrinks.
Universal high income isn’t a charity program. It’s the calculated dividend paid to the remaining humans who inherited a fully automated planet.
The companies that survive the next twenty years won’t be the ones fighting to retain human talent.
They’ll be the ones that transitioned their execution to a synthetic workforce before the demographic floor dropped out from under them.
The AI race was never about competitive advantage.
It’s about plugging the gap before the entire economic model collapses from a shortage of people to run it.
English

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/nanoc…
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.

English

It’s absolutely perfect…
A liberal man in NYC yelling about how everyone is welcome, as a Muslim man yells "Allahu Akbar!" and uses him as a literal springboard to throw a homemade bomb.
The West’s suicidal empathy on video.
Walter Masterson@waltermasterson
I was in the middle of saying “as a born and raised New Yorker, we welcome everyone into this city” when he threw that over my head.
English












