alan-stark

Thank you! Parr, Pezzulo & Friston looks like the kind of book I'm after - love cross-disciplinary works. After all, knowledge and nature are continuous. It's humans who like to chop them up into subfields :)

Tangentially related: It appears that TD ideas pop up in diffusion models, VAEs and neural net training dynamics. Any author/reading advice on links between thermodynamics, information, and neural nets?

Yep, it's the early days. Eventually we'll work out something like Design Patterns for Hybrid Development, where humans are responsible for software architecture, breaking requirements into maintainable SOLID components, and defining pass/fail criteria. Armed with that, LLMs will do the actual boilerplate implementation and serve as our Rubber Ducky Council for Brainstorming :)

Agree. We've seen cowboy developers who move fast by producing unreadable code and cutting every corner. And sometimes that's ok. Say you want a proof of concept to validate demand and iterate on feedback. But we want maintainable and reliable production code we can reason about and grasp quickly. Tech debt has a price to pay and looks like LLM abusers are on a path to waking up with a heavy hangover :)

But why would you want to review long AI PRs in the first place? Why don't we apply the same standards we apply to humans? Doesn't matter if it was AI-generated, outsourced to Upwork freelancers or handcrafted in Notepad. Either submit well-structured, modular, readable, well-tested code or PR gets rejected.

Can you elaborate on the "mode of peril"? Is it:

(a) Top labs quietly signing deals for military deployment of frontier models in unmanned strike weapons?

(b) Top labs agreeing to license LLMs for social engineering/propaganda ops?

(c) Models that vastly exceed human intelligence and have capacity to pursue own agenda (i.e. runaway intelligence)?

(d) Something else?

It looks like dangers of AGI are overblown (perhaps partially due to grant funding and ability to get political traction/investment/competitive advantage), while (a) and (b) are severely underdiscussed. Would love to get other perspectives.

At ~9:50 Michael Moritz talks how the 100-year media history tells us that if you can gather a large audience in one place, you will be able to sell them advertising.

It's likely that Google will not be the last search company. What business model could a successor use to generate revenue without falling into the ad trap? Why didn't early search engines choose other models like subscription, revenue sharing with telecom companies, or turning queries into marketing signals for manufacturers/merchandisers?

While John talks about the period before Google, this quote brings up a remarkable déjà-vu:

[~14:30] Although the search companies were having great success, they were turning into something quite different than what they started out to be. They lost the sight of what brought users to them in the first place: The need to find things. The search engine companies stopped caring about search.

When it came to actually locating relevant information on the web, Yahoo, Excite, and the rest of the so called search companies, frankly, stunk. You could spend all day typing various combinations of keywords that you were looking for. Most of the results were links to sites trying to sell you something you didn't want. The world was hungry for a radically better way of searching the web.

Reading these answers reminded me why I love HN - actually thoughtful perspectives :) Guess a lot boils down to two variables - (a) suggestion UX quality and (b) definition of 'rejection' event. I skimmed through the paper and it turns out that 91% figure is based on feedback from 70 people and anonymous feedback wasn't allowed. So, 'overwhelming 91% favorable' can be paraphrased to `64 people out of the total 16k user base said they liked it'. Would be interesting to see indirect metrics like retention on day 15.

The abstract says ..we present metrics from our large-scale deployment of CodeCompose that shows its impact on Meta's internal code authoring experience over a 15-day time window, where 4.5 million suggestions were made by CodeCompose. Quantitative metrics reveal that (i) CodeCompose has an acceptance rate of 22% across several languages, and (ii) 8% of the code typed by users of CodeCompose is through accepting code suggestions from CodeCompose. Qualitative feedback indicates an overwhelming 91.5% positive reception for CodeCompose.

In other terms, out of 4.5 million suggestions about 80% were off, yet there is 91% positive reception. That's 3.6 million rejected suggestions that potentially distracted programmers from doing their work. Yet users are happy. Is there a contradiction in these figures?

What's the value of bringing Facebook/Instagram mechanics into HN? Wouldn't that skew social dynamics away from egalitarianism, giving rise to "influencers", social bubbles and rise in clickbait? I think that not having a 'follow your friends' mechanism is a feature of HN.

My impression is that many places that label themselves as hacker/maker spaces are some version of a commercial coworking space with equipment. They have the tools, but not the spirit. That's why I mentioned Noisebridge and CCL. E.g. in CCL you can join open projects like making cheese without milk. That's quite different than 'pay membership fee X, here's the equipment, do what you want'.

Do you know of specific places in Europe that capture that spirit? With both diversity and depth of knowledge in different disciplines?