Strategy | innovation | knowledge creation in science, engineering, and technology | organization | cognitive science | AI
San Francisco
- Tweets 36,502
- Following 1,905
- Followers 2,719
- Likes 39,610
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.
1
3
444
Florida leaders are building new universities in West Palm (Vanderbilt) and Miami (Carnegie Mellon) while Florida’s governor yells at clouds.
1
82
This figure on who was responsible for technological innovation is fascinating.
LLM classification of where 37k inventions listed on Wikipedia were invented. From nber.org/system/files/workin…
10
336
Brian Gordon retweeted
LLM classification of where 37k inventions listed on Wikipedia were invented. From nber.org/system/files/workin…
5
16
97
14,455
Brian Gordon retweeted
This is my guess too.
Right now everyone is addicted to subsidized compute. Companies are giving away massive amounts of inference and giving away their agents for free.
Here's my prediction:
When the magic version of RSI does not happen, the musical chairs stop for this cycle.
What do I mean by the magic version of RSI?
I mean there is no free lunch, you can't violate the laws of physics, most knowledge is not verifiable, and the subject/object problem of "how do I know I improved when I am the judge judging whether I improved?" All of that comes back to bite you and get a lot of useful improvement but not improvement everywhere across the board in every domain to infinity that makes it go FOOM and it becomes a hidden S curve like everything else in life.
Then we end up with super useful but not magical intelligent machines, nothing like what grifters and psychotics and dreamers predicted (all jobs annihilated, utopia/dystopia, a bunch of people dying (except in the way they always do, with machines of war that use the new tech in dark ways, like all tech before it, and which already exists and is already killing people in the Ukraine war and because governments won't be stopped making weapons of war by any law proposed or on the books).
So where does that leave us?
What we will end with is a very amazing, very useful, very revolutionary new world that changes fast at times and slow in others, but with AI that is flawed, spiky, imperfect and just normal-ish, as in normal like other revolutionary technological changes like the steam engine or the printing press or the Internet, meaning groundbreaking and wonderful and awful all at the same time but not the end of all things or the start of Utopia.
2
1
437
Brian Gordon retweeted
This is not necessary. Something well provisioned leaves the soil richer so that many possibilities exist. It invests in a future where it may not even be required at all.
1
2
1
129
Brian Gordon retweeted
There are models that will deplete a zone of abundant cultural richness and then sell you back thin substitutions that lock in dependence to future rents, extracting all the while, and they’ll call it heroic innovation.
It’s not “early days” it’s enshittification.
1
2
3
296
Non-question begging question: you write a paper with colleagues who know things you never will and who can do science things you will never be able to accomplish. You are accorded partial credit for (and epistemic responsibility over) the corporate whole. 1/n
1
106
Note, the PP doesn’t say in the team case that any individual could do the entire project; it is the more modest claim that the contribution you make in the reciprocal and loopy generative processes that lead to a paper are the same across the divide. 5/n
1
36
Brian Gordon retweeted
I’ve certainly marked enough undergraduate essays to conclude that thinking isn’t central to writing
56
78
1,357
36,101
The new DeepMind paper on AI consciousness cadences looks very interesting. Two immediate thoughts: 1) two orders of magnitude difference depending on priors is huge and 2) per 👇I am just not sure we can compose credences this way. Looking forward to reading.
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.
1
125
Brian Gordon retweeted
Wake me up when we get to Artificial Intuition.
Until then, we haven't bridged the uncanny valley of the mind.
5
1
10
623
Brian Gordon retweeted
This is pretty interesting and fits in various ways with how @mattbeane has been thinking through likely shifts via and around these technologies - that is, that they will ultimately lead to increased quality in most spaces.
Folks tend to imagine that norms, including professional norms, just disappear, but, in fact, tool users *get normed* by people around them. How it's always worked.
This is a great hot take. Why hasn't econ been taken over by AI slop? Because field norms mean many publish ~1 paper per year and high quantity in low-ranked journals is seen negatively even pre-AI. Hence AI can lead to quality upgrading. This sounds right to me.
2
2
6
560
The gating function driving this norm is journal bandwidth vs tenure and hiring criteria. If journal capacity is constant, the arms race is nominal quality, not quantity. Which means this isn’t related to knowledge creation at all really, just the job market tournament.
This is a great hot take. Why hasn't econ been taken over by AI slop? Because field norms mean many publish ~1 paper per year and high quantity in low-ranked journals is seen negatively even pre-AI. Hence AI can lead to quality upgrading. This sounds right to me.
1
8
513
