Strategy | innovation | knowledge creation in science, engineering, and technology | organization | cognitive science | AI
San Francisco
Joined July 2011
- Tweets36.9K
- Following1.9K
- Followers2.7K
- Likes39.9K
Pinned Tweet
Let me try and articulate what is see is a genuine tension between Bayesian subjective credences across theories of the world and what constitutes a probability calculus within a possible world bundle.
Brian Gordon retweeted
me: kinda fucked up how many people are offloading their whole lives to muse or instinct or grokbot or whatever
claude: yeah
Without being a lawyer or knowing anything specifically relevant to the S-1, I would still suggest that it’s naive to think law just applies as opposed to evolving to meet the changing technological landscape. Reading Mary Parker Follett and Demsetz would strongly suggest this.
The most important risk factor for Anthropic is the rule of law actually being applied.
"Any determination that we are liable under such theories, or any judicial or regulatory expansion of tort, product liability, or consumer protection law, or criminal liability, as applied to AI models, could result in significant liability and require us to change how we design and offer our models, platform, and applications."
Brian Gordon retweeted
Kevin: I’m struggling with ongoing claims about independent AI discovery.
This looks more like AI free riding (intentionally or not) on human lines of attack: Navier–Stokes, now Yang–Mills. Worth checking what folks working on these problems say, e.g. xitter.cf/__alpoge__/status/2114…
Brian Gordon retweeted
The point of intellectual work is not just to arrive at answers no one else has found yet. The process of working through something can change the questions we ask, and open up possibilities we hadn’t even considered. Why would AI reaching the same conclusions make that any less worthwhile? Poverty mindset.
Wisdom of the crowd effects work by aggregating dispersed information into signal about underlying features of the world clouded by (some kinds of) uncertainty. They also incentivize strategic misinformation provision and mkt manipulation, degrading signals & epistemics.
Brian Gordon retweeted
Because equity has unlimited upside, these pitches attract a more compliant kind of capital than debt markets, where losses loom far larger. Ultimately governance seeks to be moving to debt mkts, when, where, and if ‘go/no go’ funding decisions arise. So pointillist governance?
There are task classes where this seems like it’s the inevitable equilibrium.
wrote about AI AS THE INFINITE WORK MACHINE, creating a world where we're all just constantly deploying agents to fight against everyone else's agents, running constantly to end up in the same place:
bloomberg.com/news/newslette…
Tempered by the fact that agents just aren’t great at this (at least yet).
Brian Gordon retweeted
Surprised this isn’t a more standard take in some circles. It’s great that text data had a ton of built in complexity we didn’t know would be there. But for bio, we almost certainly need qualitatively new ways to measure things. New instruments etc.
Brian Gordon retweeted
It would be shocking if say the total surplus appropriated by Alphabet under an alternative ‘secrecy and max appropriability’ policy was even a fraction of what Alphabet alone has already reaped under the OSS/academic norm policy regime.
This decision was downstream of a commitment to open source and academic norms around research. And arguably, it vindicates both counterfactually, so far at least: faster adoption, more competitive mkt, more innovation around the core technical advance, greater total surplus.
Brian Gordon retweeted
Another thing: AlphaFold is like a tiny, tiny, slice of all of biology. It's a great technical accomplishment, but it applies to static, well-formed structures (not intrinsically disordered proteins - roughly 30% of the entire proteome), and still cannot handle dynamic multimers well. It also depended on having enough data: 200,000 structures in the PDB and counting.
Saying that AlphaFold works, therefore we could "solve" biology with enough compute, is like saying that because we can predict the shape of every component in a car, we can predict everything the car will do on the road. And saying that compute or reasoning can somehow magically compensate for data is a religious belief.
If you had an exponential amount of compute you could solve simulate biology exactly and wouldn't need to run any experiments.
Some of you will think "of course, but that's an unrealistic assumption" and others will think "no, you need experiments even with that much compute. That's just How It's Done".
You would of course need evaluations once you have a solution, but you don't in principle need experiments to find it: Experiments are a way to save compute, dating from a time when it was much more expensive.
And I think the presentation is exponential but real systems aren't irreducibly exponential. AlphaFold worked, after all.