ijk

16 ms·

181.
by ijk 1y ago
A difficult, but not intractable problem: OLMoTrace claims to be able to trace from output to training data in seconds [1]. Notably, it can do this because OLMo itself was intentionally designed to be open and transparent [2]; it was traine
182.
by ijk 1y ago
You'd think that, but it's a common enough phenomenon that satirical articles joke about it [1]. [1] https://hard-drive.net/hd/video-games/huge-earthbound-fan-ex...
183.
by ijk 1y ago
There were initial difficulties in finetuning that made it less appealing early on, and that's snowballed a bit into having more of a focus on RAG. Some of the issues still exist, of course: * Finetuning takes time and compute; for one
184.
by ijk 1y ago
There's some precedent for that: you can do some useful things with the cross entropy of the two models. And k-fold cross validation might also be relevant.
185.
by ijk 1y ago
> it might just up and decide to include numbers outside of that range, or (with a nonzero probability) it might decide to list animals instead of numbers For example, all of the replies I've gotten that are formatted as "Here
186.
by ijk 1y ago
True, though in practice speed optimizations and instabilities on the GPU often lead to LLMs being very non-determanistic in practice. Which doesn't detract from your main point: there's not a lot of distinction between hallucinat
187.
by ijk 1y ago
April 4 was, notably, before the immigration issue conflict hit a flashpoint: NYT (Apr 23, 2025): "Trump’s Approval Rating Has Been Falling Steadily, Polling Average Shows" https://www.nytimes.com/2025/04&#x2F
188.
by ijk 1y ago
I do wonder if recursion is particularly hard for LLMs, given that they have a hard limit on how much they can loop for a given token. (Absent beam search, reasoning models, and other trickery.)
189.
by ijk 1y ago
There's local applications of parallel processing; your average chatbot wouldn't use it, but a research bot with multiple simultaneous queries will, for example. Better local beamsearch would be really nice to have, though.
190.
by ijk 1y ago
Inference using an API costs money. Not a lot of money, per million tokens, but it adds up if you have a lot of tokens...and some of the obvious game uses really chew through the tokens. Like chatting with a character, or having the NPC cha
191.
by ijk 1y ago
I think React is one of those areas where consistency is more important than individual decisions. With a lot of front-end webdev there's many right answers, but they're only right if they are aligned with the other design decisio
192.
by ijk 1y ago
There's two general categories of local inference: - You're running a personal hosted instance. Good for experimentation and personal use; though there's a tradeoff on renting a cloud server. - You want to run LLM inference o
193.
by ijk 2y ago
Alas, current LLM prompting has a lot of hacks. Half of them are useless, of course, while the other half are critical for success. The trick is: which one is which?
194.
by ijk 2y ago
One way I've heard it summarized: Computer Science as a field is used to things being like physics or chemistry, but we've suddenly encountered something that behaves more like biology.
195.
by ijk 2y ago
I didn't really expect that we'd find a potential treatment, like, ever. Hopefully the problem is actually as tractable as the animal models here suggest.
196.
by ijk 2y ago
I'm curious if you find their new website interface more tractable--there's some inherent friction to the prompting in either case, but I'd like to know if the Discord chat interface can be overcome by using a different inter
197.
by ijk 2y ago
That pretty much sums up the objection from EFF: as written, it would ban both publishing research and reading research for pretty much all of AI. It's unlikely that this will ever get out of committee, of course.
198.
by ijk 2y ago
This would assume that Hawley knows the difference, or that the court will care after the carelessly written legislation passes. (It's unlikely to ever get close to passing, but as the changes to R&D expenditure rules and the TikTo
199.
by ijk 2y ago
On the one hand Hawley proposes a lot of things that don't ever get close to becoming law. On the other hand, we did ban TikTok (which is currently unavailable on the app stores because of the ban). I can think of few ways to more effe
200.
by ijk 2y ago
I'd assume that the existing llama.cpp ability to split layers out to the GPU still applies, so you could have some fraction in VRAM and speed up those layers. The memory bandwidth might be an issue, and it would be a pretty small perc
201.
by ijk 2y ago
Most AI generated images are like most dreams: meaningful to you but not something other people have much interest it. Once you have people sorting through them, editing them, and so on the curation adds enough additional interest...and for
202.
by ijk 2y ago
The problem with "provide LLM output as a service," which is more or less the best case scenario for the ChatGPT listicles that clutter my feed, is that if I wanted an LLM result...I could have just asked the LLM. There's may
203.
by ijk 2y ago
This would explain the unexpected benefits of training on code.
204.
by ijk 2y ago
I've seen two approaches so far: Grounding everything in symbolic representations. [1] Which can greatly empower stuff that we could simulate but was too complicated to write a game around; now you can have agents respond to complex si
205.
by ijk 2y ago
We already do this for songs; anyone can pay the mechanical rate and record their own cover of a song. It is an imperfect comparison, since a cover is its own recording, and ongoing royalties are involved, but the point is that there are so
206.
by ijk 2y ago
If you track down the original source, there's a video of the lead author of the Nature paper explaining how it works: https://www.caltech.edu/about/news/gargantuan-black-hole-jet...
207.
by ijk 2y ago
In creative writing the problem becomes things like word choice and implications that have unexpected deviations from its expectations. It can get really obvious when it's repeatedly using clichés. Both in repeated phrases and in tryi
208.
by ijk 2y ago
It's not quite that straightforward; since close kin also share genetic heritage anything you do to benefit your near kin also propagates some percentage of your own DNA. Kin selection has been part of evolutionary theory since Darwin:
209.
by ijk 2y ago
From the abstract: "Across settings, we find a consistent results that code is a critical building block for generalization far beyond coding tasks and improvements to code quality have an outsized impact across all tasks. In particula
210.
Exploring Impact of Code in Pre-Training(arxiv.org)
5 points by ijk 2y ago|2 comments

More ›