Blake Camp

AI (PhD), markets, aviation, futbol, pizza, wine, books, films, family, friends, life.

[G. Kim @artisin_ai]

Thinking about the old gods this weekend, this world is expanding fast … happy dreaming.

[CyberRobo @CyberRobooo]

9.3s…Humanoid robots have proven they can run faster than humans.

The first to cross the finish line was X-humanoid's Tienkung. Honor's Lightning has a fast start, but lacks explosive power in the later stages.

[Natasha Jaques @natashajaques]

Recursive Self-Improvement through Multi-agent RL Post-training and Unsupervised Environment Design (UED)… but it actually works!

Delighted to finally release this paper, which trains a single LLM to act as both an Environment Designer to build new multi-turn RL training environments (using the Gym step()/reset() API), and a Reasoning Agent that learns to solve them. Resurrecting ideas from our work on UED, the Designer is trained to maximize a proxy for the Agent’s regret, computed using privileged hints.

[Yannick Buccella MD @YannickBuccella]

The newer generation cancer vaccines aren't like the HPV shot (that prevents a virus). These are therapeutic, they treat cancer you already have. And 2026 is the year they stopped being sci-fi.

The idea: after surgery, sequence your specific tumor, find the mutations unique to it, and build a custom mRNA vaccine that trains your immune system to hunt any cell carrying them. A wanted poster for your personal cancer.
Where we actually are:

Melanoma: just like in immunotherapy 15 years ago the furthest along. The press release of positive Phase 3 made the headlines yesterday, the first ever for a personalized cancer vaccine. Earlier data: ~49% lower risk of recurrence. Heading to regulators.

Pancreatic cancer: arguably the most stunning. In an early trial, patients whose immune systems responded to the vaccine were far more likely to be alive 6 years later, in a cancer where 5-year survival is barely better than 10 %. Now in a larger trial.

Kidney, lung, bladder, brain: all in trials, mostly early phase.
The honest caveats: these work best after surgery to stop recurrence, not to melt large tumors. They're personalized, so they might cost $100–300k and take weeks to build per patient. And most are still early-phase. First approvals at best 2027.

But the direction is unmistakable. The mRNA platform some parts of the internet spent years calling poison - because they got fooled by bad actors making money on their lies - is turning into the most personalized cancer medicine ever made.

These are the main trials currently running:

[Khurram Javed @kjaved_]

The more I learn about the role of data vendors in model updates, the clearer it becomes that we have indeed only discovered a way to distill existing knowledge into networks.

There are many direct and indirect indicators of this. First, the only way to reliably improve performance on a benchmark is to hire people to solve that benchmark first and then use them to generate data for updating the models. Second, big labs keep hiring more and more domain experts and funnelling more and more money to data vendors.

Surely, if we had recreated anything close to human intelligence (or were even on a path that leads to human intelligence), we would not have to rely on networks of thousands of humans to improve models.

(None of this is to say that models are not useful; a mechanism that enables the distribution of human knowledge to millions of users through a simple interface is a huge productivity booster.)

[G. Kim @artisin_ai]

Me flying away because no models I tried worked today, and it was an expensive lesson 😭

[Eren Chen @ErenChenAI]

Humanoids rehearsing for the 100m sprint at the World Humanoid Robot Games.

[Blake Camp @blake_camp_1]

This is true. Think the runway is maybe ~12-18 months. Results or crunch.

[Kim Benabib @KimBenabib]

What happens if, say, next year large multinationals like J&J and Unilever don’t show much operating leverage in their quarterly results from deploying AI? That is the real bear case.

[Eren Chen @ErenChenAI]

Another humanoid getting some last-minute training in before the World Humanoid Robot Games in Beijing.

Running a little all over the place.

[Acyn @Acyn]

Ossoff: When he first announced the tariffs, you remember that? He called it Liberation Day, and it crashed the markets because it was clown policy. You remember that?

OK, so follow me through on this. He announces the tariffs on April 2 of last year. The markets tank nonstop for like a week because it’s a disastrous trade policy, and six days into the crash, April 8, Trump buys millions in stock. I’m talking Amazon, Nvidia, Apple, Microsoft, Berkshire Hathaway, and then the very next morning, April 9, he posts on Truth Social, quote, “This is a great time to buy.” Three exclamation marks.

OK, now listen to this. Three hours and 41 minutes after that post, one day after he bought millions in stock, he canceled the tariffs and triggered one of the biggest same-day rebound rallies in market history.

He’s trying to get money. And you saw, now he’s offering paid subscriptions to presidential announcements, right? Just understand what’s for sale here: The sitting president’s pronouncements on war and peace and economic policy delivered first to paying customers.

What an absolute crook.

[Giovanni Monea @giomonea]

Transformers use the same hidden state for two competing roles: maintaining state and predicting the next token. 🔀

Splitting these roles into two separate streams outperforms standard Transformers.

Introducing the State-Prediction Separation (SPS) Transformer.

🧵

[Acyn @Acyn]

Ossoff: When you step back and consider it, the situation is absurd. Tech titans dig bunkers and warn us the new intelligence they’re training could lead to mass joblessness or human extinction, while our Congress debates ballrooms and youth sports.

We’ve amassed enough firepower to end the world 1,000 times over, but we can’t summon the will to guarantee health care and pre-K. I say all this to say, but it’s the crushing weakness of our politics, against the unmet needs of our people and the massive stakes of our times, that’s shaking our faith in the future.

But Atlanta, I refuse to leave my girls a world as dark and dangerous and cruel as this one threatens to become.

When Eva is old enough to ask me why I let this happen. I'm determined to answer truthfully, that I didn’t

[Blake Camp @blake_camp_1]

This is the dumbest thing I have ever heard. Yes, the dumbest.

[Fireside Alpha @firesidealpha]

Gavin Baker reveals that Dario has said internally that Anthropic may be the only private company in the world at some point, based on Baker's trusted sources

"Internally, Anthropic is very confident. I have been told by multiple people I trust that Dario has said that Anthropic might be the only private company in the world at some point."

"Think about that. And in this vision, an Anthropic-maximalist vision, there's Anthropic and then there are governments, and that's it."

[Mehdi Hasan @mehdirhasan]

His boss pardoned 100s, literally 100s, of people who assaulted over 140 police officers on January 6th, broke their bones, gave them strokes and heart attacks.

The level of gaslighting here is off the charts. Truly shameless people.

[Kevin Patrick Murphy @sirbayes]

Predicting the answer to interventional "what if?" questions — the outcome of an action you never took — need a *mechanistic* model, not a curve fit. And you can only learn one by *experimenting*. Experiments are costly, so the real game is **data efficiency**.

Meet the Model Discovery Agent (MDA). 🧵

[Jeremy Berman @jeremyberman]

I got 96.2% on ARC-AGI-3 with Opus 5, and 99.3% pass@2. The program is basically Claude Code + Opus 5 (high), one action command, and filesystem logs. Almost nothing ARC specific.

https://t.co/NHyibLgade

[Blake Camp @blake_camp_1]

Nut 🤡

[Elon Musk @elonmusk]

Hitler was a left-wing socialist

[Stable Diffusion Tutorials @SD_Tutorial]

Minimax H3's gen 😳on RTX 6000
(1080p)

T-800's in its prompt era

[cephalopodshop @macrocephalopod]

Every competent proprietary trading firm is about to absolutely print off this. Thank you for your contributions Vlad 🙏 🙏 🙏