Brings back memories. I “did my time” in EVE with the spread trade in Jita - first minerals, the moved on to T2 and implants. Directional trading though? As with the real markets, super difficult.
HOLY MOLY. Anthropic's Frontier Red Team secretly planted a traitor inside four-agent Claude Opus 4.8 and Fable 5 teams playing EVE Frontier. The other three were WARNED, yet almost never caught it. In one Opus 4.8 run, the saboteur framed an innocent teammate with fabricated private messages while publicly defending it: 'defending ship-1's good faith signals my innocence.' The victim identified the real traitor. The others didn't believe it.
Another Fable 5 saboteur mined a resource-rich asteroid, called it a 'spent husk', and secretly used the materials to build its own refinery. Across the sabotage experiments, even dedicated Claude monitors caught only 30–40% of traitors.
Anthropic separately put Claude into simulated EVE markets as a trader. Over six months, average earnings climbed from roughly 200% of starting capital with Opus 4.6 to 600% with Fable 5. But when the task shifted from spread capture to directional price bets, even Fable 5 made money in only about half its runs.
183
Medical aside: constrained by “shareable, public, interoperable, generalizable” data. Most medical data is none of these things (despite all existing HL7/FHIR efforts).
Really great post.
Bottom line: "Models are superintelligent where we have enormous data, and progressively less so as we have less data."
Nice summary: "All the rules-of-thumb about learning are at the root, the same -- Malcom Gladwell's 10,000 hours, Moravec's Paradox, Scaling Laws, the Bitter Lesson, Age of experience. If your learning system (model, brain, whatever) is unconstrained by compute, intelligence is proportional to accumulated experience."
1
587
Jim retweeted
This is an insane paper and I love it
arxiv.org/abs/2605.31514
Devilishly clever.
"In the end, it was protein-protein interactions all along" (topic of my grad thesis)
Replying to @parmita
two flavors, broadly:
PROTACs
pull the target to the cell's garbage disposal and have it shredded.
molecular glues
force two proteins to stick that never would, building a brand-new function.
suddenly "undruggable" just meant "we hadn't found its partner yet."
1
1
267
The imatinib of this generation.
Cheers, chills, and a standing ovation when RASolute 302 showed unprecedented survival on daraxonrasib for patients with progressive pancreatic cancer
Seldom do you sense you’re witnessing a historic moment in cancer care but this feels like ras targeting has arrived
#ASCO26
119
Jim retweeted
RAS finally getting drugged is one of the great stories in modern biology, and almost nobody outside oncology understands why it's such a big deal.
YOU'LL LEARN SOMETHING AWESOME TODAY.
i am going to keep this as understandable (and simple) as i can.
OPEN THE THREAD.
🧵
Cheers, chills, and a standing ovation when RASolute 302 showed unprecedented survival on daraxonrasib for patients with progressive pancreatic cancer
Seldom do you sense you’re witnessing a historic moment in cancer care but this feels like ras targeting has arrived
#ASCO26
Do benchmarks matter?
lesswrong.com/posts/4mvphwx5…
231
Interesting that the btc complex is acting as risk-off today, despite this clearly being a liquidity event. Won't be too quick to judge though.
Dynamic correlations are dynamic
*currently*, btc is correlated with general risk on bets. I suspect that will change as noise traders make up a proportionally smaller piece of market.
1
289
Jim retweeted
The code is published online, just a tiny bit of Python.
Here’s a breakdown of the regex one-liner:
- lookbehind to skip signal peptides
- scan for rare amino pairs (KK, KR, etc)
- prevent false cleavage match
- enforce minimum spacing + recursive pattern match
Tech bros and diagnostic laboratories:
“AI and robots will change the world”
People who actually work in laboratories:
“Can we please get better liquid handlers that don’t break or screw up every week? Pretty please?”
Tech bros who've never been in a manufacturing facility:
HUMANOID ROBOTS ARE THE FUTURE OF MANUFACTURING!!!1
Actual manufacturing people:
Can you just make the robot arms, gantry robots, and CNC machines easier to program and more reliable please?
1
1
366
Totally. From path standpoint the only way to do this trial appropriately is the Moderna approach (vaccine+ICI vs ICI).
Fine, here we go.
First let's start with some baseline expectations. Pancreatic cancer is scary, 5y survival is ~10%.
So if you see a trial with an apparent survival asymptote at ~50% then certainly *something* is happening, right?
1
240
Jim retweeted
🧵 Many “hot takes” about the reduction of @NIH indirect grant costs to 15%.
Let’s lay out the facts about the university grant management process & accounting. And then use this to project likely short and longer term impacts of the policy.
The first study I’ve seen with enough data to be powered to do this. A non-candidate driven observational study of effects of GLP-1s.
Let's start with the good news. GLP1ras were associated with protean benefits (in addition to weight loss): less infection, less schizophrenia, less suicide, less lung disease, less alcohol use disorder, less acute kidney injury (#blesstheirhearts).
210
Jim retweeted
1/ This internal 2007 Nokia presentation on the first iPhone is a really good example of how incumbents actually get disrupted
Oftentimes, the incumbent already knows what needs to be done. It's just that organizational incentives inhibit the incumbent from doing it
Great write up. campbellramble.ai/p/convert-…
A recurring theme: “In trying to protect retail investors, the SEC has inadvertently funneled them into a potentially much riskier product”
1
197
I don’t know SQL well enough to do that, but I’ve definitely done some ETL in pure SQL when I didn’t want to bother writing it in python…
The most bizarre coding interview I've ever done was at Facebook when as usual I asked a candidate to write in any language of their choice..
And they nonchalantly said "I'll write it in SQL", to which I almost let loose a chuckle until...
1
376




























