Jon Charbonneau 🇺🇸 retweeted
MAYBE I WAS WRONG ABOUT BLOCKCHAIN TECHNOLOGY
I wrote about how I always used to make fun of the guys who said things like "Bitcoin is speculative, but the underlying blockchain technology is interesting." And now I think maybe they kinda had a point.
gigantic breakthrough if proved out
very long but must read
very TLDR - mechanistic interpretability is likely way more tractable than previously thought because past assumption that AI models jumble unrelated ideas together in neurons (superposition) was probably mostly wrong
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had
with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works
already-clean.goodfire.com/
Jon Charbonneau 🇺🇸 retweeted
Hyperliquid is unlike any other crypto endeavor I can think of.
It’s not another memecoin casino. It’s a focused endeavor to re-architect finance and it’s the one where people are trading oil perps 24/7. It’s not a household name, but it’s caught up to Sol in USDC issuance
NEW ODD LOTS:
It’s the podcast debut for Jeff Yan
@tracyalloway and I have an ultra-rare interview with the CEO and co-founder of Hyperliquid Labs about his plan to house the world’s entire financial system
open.spotify.com/episode/5Aq…
Jon Charbonneau 🇺🇸 retweeted
i think this flow for raising capital is pretty underrated.
lots of apps in crypto are kind of hard to back as just an idea. what if you can raise a small runway (6 months or so) from public and test out idea first?
if failed, liquidate and shut it down. if successful, raise from funds and angels at a markup to public ico to fully build it out.
We have completed OTC sales of $CARS to DBA, Novora, and Anas Khader.
All three transactions were executed at USD $0.1065 per token, with no vesting and no lockup.
• @dba_xyz — 2,816,901.41 $CARS — $300,000
• @novora_ — 469,483.57 $CARS — $50,000
• @allinbitcoin_ — 1,173,708.92 $CARS — $125,000
Total: 4,460,093.90 $CARS for $475,000. All proceeds will go to the treasury.
Jon Charbonneau 🇺🇸 retweeted
Welcome to Backable, a global fundraising platform for frontier ideas built by MetaDAO, designed for the bold, the brave, and the radically different.
Build a business. Chase a dream. Be early to something big.
The only question that matters is “Are you Backable?”
Jon Charbonneau 🇺🇸 retweeted
We have completed an additional $350,000 OTC sale of solana:CARSsxWPkpQWvfyRBwfGMGvysJBHdHGfE46X5MNgmeta to @Colosseum.
The transaction was executed at USD $0.0987 per token for 3,546,099.30 solana:CARSsxWPkpQWvfyRBwfGMGvysJBHdHGfE46X5MNgmeta, with no vesting and no lockup.
All proceeds will go to the treasury.
Thesis:
1) @allocateur & @AgentMechanic are hustlers chasing a totally untapped market that they're really excited about
2) hot wheels are awesome
Check out the app its a lot of fun
We have completed OTC sales of $CARS to DBA, Novora, and Anas Khader.
All three transactions were executed at USD $0.1065 per token, with no vesting and no lockup.
• @dba_xyz — 2,816,901.41 $CARS — $300,000
• @novora_ — 469,483.57 $CARS — $50,000
• @allinbitcoin_ — 1,173,708.92 $CARS — $125,000
Total: 4,460,093.90 $CARS for $475,000. All proceeds will go to the treasury.
The P(doom) forecasts I actually want to see:
1) prediction markets on P(doom)* for given dates
2) conditional markets showing change in P(doom) for specific actions (e.g., how much does P(doom) shift for date X if the Ban Artificial Superintelligence Act is enacted by date Y)
*P(doom) needs to be more narrowly defined than "everyone dies". Market won't function properly if nobody can collect resolution. You can more narrowly define "doom" (e.g., some large % of global population is killed) which in practice will surface mostly the same useful info.
The P(doom) forecasts I don't want to see:
xitter.cf/jon_charb/status/20984…
exactly this
random AI employees aren't experts in forecasting p(doom)
p(doom) is a thought experiment requiring broad real world understanding more than AI tech expertise
similarly this is why the AI labs have all been wrong about predicting economic impacts of AI to date
Jon Charbonneau 🇺🇸 retweeted
GPT-6 Astra is more aligned than our previous models. But it’s also less monitorable, which is a concerning trend that we take very seriously. We believe monitorability drop comes from a jump in intelligence and not direct optimization pressure on CoT or architecture changes. More thoughts in the thread.
"GPT-6 Astra shows a substantial decrease in chain-of-thought monitorability compared to previous models"
Astra has increased ability to control or shorten its CoT to evade monitors under adversarial conditions
Mech interp importance is increasing
deploymentsafety.openai.com/…
the nonsensical decel anti-growth mindset is the biggest fundamental issue with the Dems
theres ~3 big economic levers - growth, taxes, & government spending
neither party has great policy on taxes or govt spending
but being pro-growth bails you out of a lot
Jon Charbonneau 🇺🇸 retweeted
MetaDAO is rebuilding venture capital for the internet.
Check out our new website to learn more metadao.fi