Профиль

The boring company sewer tunnels are meant to give you The Wormhole Travel Experience®

Yes, of course. I used to think of it as the primary motivation, and it likely still is. But I think that that underestimates the well-meaning but dangerous utopia-dystopia imagination induced megalomania prevalent in EA adjacent safety circles. They know what's good for you better than you. :P

Over time I have come to the conclusion that a lot of this isn't as much about regulatory capture and becoming a designated economic tollbooth (although there's that too). It's more like they really believe 1) Powerful models are dangerous. 2) They're pretty much the only trustworthy stewards.

Despite all the absurd polemics and parochial (and polarizing) opinions, he had a fundamental point about the nature of math. Regardless, it's funny their "discussions" were predicated on timelines of many decades. Like, about abstract future computer mathematicians.

Given all the discussion around math, punctuated by articles by W. T. Gowers on the theme, the one person I am most curious to hear the opinion of is Owl Sowa.

This doesn't have practical applications at the moment unfortunately. But it seems like an important structural result (on the lines of how much geometric info does a very natural algeraic encoding retain), so it could eventually lead to a language that helps in practical applications.

While the Fields Medal celebrates the under-40s, the mathematician Joan Birman has, at the age of 99, solved a major open problem in representations of the Braid groups, a topic she has worked on for more than 60 years.

One could argue that its brand of B2B contract recurring revenue is much stronger i.e. compared to the cyclical advertising auction supported transactional revenue (of two of the other hyperscalers, for example). It doesn't sound cool, but it is going to be great.

It's because of this that my leaps are only timed till July 2028. Microsoft is in a sense super cheap. Trading near its lowest forward price to earnings multiple since 2023. I don't buy the AI doom talk around its brand of enterprise software.

Yes, without spanish it'd be very challenging. Hopefully, I'll manage to be conversant in Spanish by then. :)

He was also very funny and generous with his time. I was 16-17, but he didn't let me feel that. I kept in touch till around 2014. He was apparently a 1980 WPI alum, and mentioned that they will most likely allow me to switch fields and give me a scholarship for a terminal MS. He was right!

I was planning to do a solo trip in 2025 (including Potosi), but have got stuck due to visa stuff. I wanted to go there first before returning to Patagonia, but alas. Hopefully some time in the next few years.

If you have to be a hawk, talk about a coherent and rational industrial policy (going towards manufacturing) and more competition.

Too much of a top gap for too long gives fodder for the fake-hawkish talk, esp if the top open models are Chinese (doesn't matter what the complaints). It's also very fragile: easy to buy out support for regulatory capture. More competition at the "frontier" than marginal revenue would be excellent.

Just from an ecosystem health perspective, it's important for Google to be back on the "frontier" model landscape. It's true that the commoditization of LMs (from open sources) helps Google the most, but regardless, the continually large vacuum for OpenAI and Anthropic shenanigans is not good.

This is a good review. Give me a name that compresses a whole lot of history any day. PS: This cites an old article by Tom Fawcett on ROC. Just googled him to learn that he died in a freak biking accident during covid. He was the one who told me to apply to WPI (we were Orkut friends).

Like every ten years rename things so that they are intelligible to everyone? Actually in ML, where community LLM-like context cliffs are approximately 6 months, this has been happening a lot.

ROC comes from radar and originally was a pretty clear name. It seems less meaningful in ML, but the idea is the same. If you think of a ML classifier as a radar receiver, both ROC and AUC are pretty meaningful. Just using older terminology is better than the current trend of spawning names.

No I won't. I like the odyssey, but I am not sure I care too much about "artistic expressions" around it. Feels like a waste of time to me.

I have def stopped talking to most of my "mutuals" there who were always encouraging these kind of things from the side (based on the fashion of the day). At the time it seemed like they were about academia and stuff, but they were attaching to bad thought patterns. I don't want to know them.