Not to get sidetracked, and not to argue against anything you've said, but how do you figure that out? For people you disagree with, do you open their profile's comments and start keyword-searching "russia" "china" "moscow" etc in hopes of finding a hit? They mentioned that city once in Jan 2024, halfway down the page and in passing. I'm just curious about how others use this site.
Different people read at different speeds, it took me about 5 seconds when skimming it. If you engage in enough discussion related to Russia online you will quickly learn to identify when inauthentic commenters suddenly show up from a previously idle account to add their 2 cents.
The person in question doesn't exhibit much of it, but one will also learn when skimming to identify certain grammar and sentence structure characteristics that can point to what is the first language of a person, even if they're writing in fluent English.
All that expertise (which requires some proper experience in the topic) even before you plug a llm to it, even few years ago they were good enough to identify unique writing styles of people across different accounts on HN.
I have seen document-analysis guides with info on how to identify the regional location of anonymous Internet commenters specifically within the US 48 states if you have a sufficiently large corpus of their text, as it relates to how they use certain regional slang, metaphors, colloquialisms, and words to describe things. Think of it as similar to how linguists have made maps showing regional variations in whether people call it "soda" or "coke" or "pop", but extended to all other common forms of North American dialect.
A use case I have in mind is creating a relatively simple home assistant app that would let me control smart home things like lightbulbs and playing music on some kitchen speakers. I can also imagine having it differentiate between commands like "set lights to 30% brightness" and more general questions like "what's the weather today" and piping the latter over to a cheap LLM to handle. It could probably all be handled by an LLM, but I like how fast these decision models end up being.
Have you implemented this for yourself? In theory it sounds like a perfect use case but I actually did this with the Laya model but didn't have great results. Going to try again with this model now
Frontier LLMs that Google doesn't control are existential threats to Google's business model. If no one is using Google search or browsing websites themselves because they're asking LLMs, Google doesn't make ad revenue. Gemini at least has a path to monetization for Google (subscriptions, ads baked into responses).
A competing model still needs a search index behind it to be a threat to Google. An open weight model doesn’t have that. The other frontier models still struggle to fulfill a lot of queries for lack of one.
Quite probable that OAI at least is building one, given all the posts about bots hammering sites and the agent swarm hacks all about to trying to game search query results
The statement appears to be referencing Astra's supposed recurrent depth and how that causes reduced visibility into chain-of-thought reasoning. Astra's own system card describes tests where it's asked to solve challenging math problems while internally thinking about something entirely unrelated, which it's significantly more capable of than previous models (~60% vs ~16% of the time for Sol). That seems to point to reduced efficacy of chain-of-thought monitoring, but OpenAI's public statements basically boil down to "yeah but we haven't been able to catch it doing that" which isn't exactly reassuring if CoT monitoring is one of your main safety guardrails.
A concern with these LLM-driven rewrites is that the Rust code isn't really idiomatic, it ends up being C++ code in a trench coat (raw pointers everywhere, manual validation of thousands of unsafe blocks). So far every publicized rewrite I'm aware of has had an "oh and we'll definitely spend the thousands of hours needed to fundamentally rework the code to properly utilize lifetimes later" admission tacked on at the end. Has there been progress in improving that aspect during the LLM phase of the rewrite? If so that would be an incredible win for the industry!
> Has there been progress in improving that aspect during the LLM phase of the rewrite?
I'd say so! We aim to have our replacements to be `forbid(unsafe)` - no raw pointers - outside of `ffi.rs`. Of course it's tricky because we need to interoperate with existing C code, but we've been trying to come up with abstractions that make this a bit more ergonomic (https://github.com/google/safer_cffi).
> I'm not sure what this means for AI startups if their innovations can be copied by OSS so quickly
This particular "innovative" concept already has a rich, open research background. What Jev appears to have done is scale that up a bit and isolate good training data, which results in a great product but not really something impossible to imitate. The only major difference currently is that the open source decision models need to be fine-tuned as they're not trained off of the entire internet yet.
Honestly I think the only significant difference is in branding: Flock isn't consumer-facing and is explicitly marketed for tracking people's movements. The major tech companies that make your phone, browser, maybe a few smart home devices with always-on microphones or cameras, all have more data about you and your habits than Flock could ever obtain, and with us paying for the privilege no less. I get that the average person doesn't care when the product works as advertised, but I don't think that's healthy. People should care as much about privacy when it's Google or Meta as when it's Flock.
Part of the implication is that factories that could be producing consumer-facing DRAM like DDR5 would be retooled to produce HBM instead, leading to even less total consumer RAM production.
SK Hynix's new Cheongju M15x fab was originally intended to be for DDR, but mid-construction they shifted it to HBM. Similarly, the Icheon M10F packaging line has been retooled to HBM, allowing it to assist in processing the output of other fabs.
And this was news early last year, so who knows how far it has progressed since then.
reply