Log In
Search
In this episode of The Econoclasts, Yanis Varoufakis and Wolfgang Munchau expose the dangerous flaws of a newly hyped Trump-Iran pre-deal, reveal how Europe’s aggressive shift toward wasteful defense spending is a delusion built on exaggerated threats, and uncover a widening technological chasm where aggressive US nationalist regulations and a stagnant European mindset are combining to leave the West severely outpaced by China’s productivity-boosting integration of AI and robotics.
I skipped forward past the first half, which it quickly became clear was a yanisrant (if this word is not in the OED yet, I am happy to write in, making the case for the OED to add it).
Anyway on to the second half about Anthropic and Fable, a slightly kneecapped Mythos-class model, which I used heavily for the three days it was available before Trump forced Anthropic to withdraw it.
I won’t comment on what Yanis said, except to note that I’m sure Yanis was wearing an invisible hat, because how else could he possibly have spent talking through it the entire second half of the episode otherwise?
I will address what Wolfgang said, and I urge him not to look at what’s happening with cynical comments about Anthropic marketing (which are besides the point), or just looking through a single politico-economic lens, because this is big.
I can best illustrate, by relating my own experience of those three days and the after-effects, and also point to other places to look.
A little work related story. We had a code feature/fix which was pulled last minute from our last release, because some side-effects came to light while testing, which we couldn’t solve simply (and we’re not neophytes).
Anyway, Fable comes along, and I thought, lemme check this out.
So I pointed Fable to the Jira, and wrote a brief instruction, explaining what I was looking for.
Fable started by saying:
Let me read the Jira…
drrr..dth.dth.dth…drr..dth.dth..
Ok, now let me read the code in the repo and the solution attempted…
drrr..dth.dth.dth…dth..dth.dth.dth..
Now I have the full picture…
drrr..dth.dth.dth…drr..dth.dth.dth
drrr..dth.dth.dth…
Writing solution…
drrr..dth.dth.dth..
Fixed and tested.
Code here… and test results here…
And that was it. Eight minutes, one solution, served warm. And me? I spent the next two hours looking through the analysis, the code created, and the tests and results, and re-verifying the results for myself. I’m no longer coding, I’ve been turned into a glorified tester! (this is not what I signed up for).
Another person with a similar story (and there are lots everywhere on YouTube, assuming Wolfgang passes the age verification checks):
https://youtu.be/zaTAtMBfMkw?is=4SVwgrc5I6NTD7Ok
I also strongly recommend, Zvi Mowshowitz’s substack if you want a clear picture of what’s going on.
And me? The withdrawal symptoms are terrible – I am like a hobbit who has found and then lost the Ring – gollum gollum.
The bigger question is: we have gone from GPT-3.5 to Mythos in three years flat. What happens in another three years?
Great comment (from someone who actually works with AI). Thanks.
I was pleased to hear Yanis make the case for the dissolution of NATO. If Europe has nothing to fear from Russia, Iran and China, then to guard against what military threat does NATO exist?