NiloCK

22 ms·

241.
by NiloCK 1y ago
Anthropic moved from 3.5, to 3.5(new), to 3.7. They skipped 3.6 because of usage in the community, and because 3.5(newer) probably passed some threshold of awfulness. People also use 3.5.1 to refer to 3.5(new)/3.6. The remaining diffic
242.
by NiloCK 1y ago
SRS is wonderful. The improvements in the linked article are real and laudable. But they are comparatively narrow and technical compared to the improvements that need to be made in usability and scalability (of content dissemination and nav
243.
by NiloCK 1y ago
Wow, thanks! The app's been 'in production' for myself and my daughters for years, but no public access or polished working demos yet. I hope to have exemplar static-site builds using very soon. I'm also working on these
244.
by NiloCK 1y ago
> Half the benefit of SRS comes from working out what the flashcards are. You have to circle around a concept, look for similarities, differences, examples, generalisations, properties, etc. This is at least half true, but among SRS folk
245.
by NiloCK 1y ago
With respect to model access and deployment pipelines, I assume there are some inside tracks, privileged accesses, and staged roll-outs here and there. Something that could be answered, but is unlikely to be answered: What was the level of
246.
by NiloCK 1y ago
What? No they aren't. You get different results each time because of variation in seed values + non-zero 'temperatures' - eg, configured randomness. Pedantic point: different virtualized implementations can produce differen
247.
by NiloCK 1y ago
Warning: you may have become cynical. I didn't read that comment as snarky at all - efficiency comparisons between emerging tech and SOTA (grass, trees) are extremely relevant! (Warning to welf: you may be naive)
248.
by NiloCK 1y ago
Slightly shameless, but easier than typing a longer reply. https://www.paritybits.me/think-toggles-are-dumb/ https://nilock.github.io/autothink/ LLMs with broad contextual capabilities shouldn&#x2
249.
by NiloCK 1y ago
I could be misinterpreting your claim here, but I'll point out that LLM weights don't literally encode the entirety of the training data set.
250.
by NiloCK 1y ago
Some info on the attacks themselves: https://rekt.news/indexed-finance-rekt https://rekt.news/mango-markets-rekt
251.
by NiloCK 1y ago
Leveraging OpenAI for a Worldcoin proof-of-humanity toehold? Blue checks for WorldID holders, and a presumption of dead internet everywhere else.
252.
by NiloCK 2y ago
Google historically allowed employees to self-direct 20% of their working time (onto any google project I think).
253.
by NiloCK 2y ago
> Aren't students already allowed to fail? It's technically allowed on an individual basis, but the economics don't work for any institution to attempt to raise its bar. If institutions X and Y grant credential Z, and X st
254.
by NiloCK 2y ago
At the risk of piling on (others have commented), I'll go so far as to say that there really is more information here if you make the gentle assumption of a human observer. In the animated version, a human observer here is allowed (for
255.
by NiloCK 2y ago
The 'before' image passes the test this time in a "Treachery of Images" sort of way.
256.
by NiloCK 2y ago
I still don't really get this. Do you mean that frent-ends have proprietary code? Contracts on-chain can be slightly inscrutable in their bytecode format, but it's pretty uncommon for smart contracts to not be published with sourc
257.
by NiloCK 2y ago
I'm not generally inclined toward the "they are cheating cheaters" mindset, but I'll point out that fine tuning is not the same as retraining. It can be done cheaply and quickly. Models getting 5X better at things all th
258.
by NiloCK 2y ago
Looks really neat on first glance, thanks for sharing. I'm working on some browser based piano learning software. Is this open source?
259.
by NiloCK 2y ago
I don't agree with this, or maybe I don't get it. The point of DRY isn't to save the time on typing - it's to retain a single source of truth. If you've used an LLM to recreate some mechanism for you system in 8 dif
260.
by NiloCK 2y ago
This is only true recently! Through most of history people lived and died with technology and culture nearly identical to that of their parents / grandparents. In terms of tech and cultural evolution, we are on the uptick of a hockey-s
261.
by NiloCK 2y ago
I am in principle in favor of inventive gameplay and twisting existing form factors, but this actually grossed me out as if it were gore and I had to close the tab before even knowing what was going on. Pretty surprising to be honest. I fee
262.
by NiloCK 2y ago
One outsized impact of this is going to be on agentic LLM programming workflows, where compile and test time are trending toward being the dominant bottlenecks. See how many spaghetti types get churned through this faster transpiler. Didn&#
263.
by NiloCK 2y ago
Breath it in, get a coffee, and sit down to solve some bigger problems. 3.7 really is astounding with the one-shots.
264.
Think Toggles Are Dumb(paritybits.me)
2 points by NiloCK 2y ago|0 comments
265.
by NiloCK 2y ago
There's a little grain of salt with respect to context lengths: the number has grown, but performance seems to degrade with larger context windows. Anecdote: I often front-load a bunch of package.jsons from a monorepo when making tooli
266.
by NiloCK 2y ago
Some particular systems have grown in complexity, but the general point isn't 'new' or emerging. No map has never been its corresponding territory.
267.
by NiloCK 2y ago
I, too, wrote a shittier version of this a little while back: https://www.paritybits.me/stack-ossification/ Another observation since then: good documentation for newer tech stacks will not save the LLM's capabi
268.
by NiloCK 2y ago
In a way I agree, but the sustainability is shaky. The intrinsic motivation for providing the comments comes from a mix of - peer interaction, comradery - reputation building If becomes evident that your outputs are only directly consumed
269.
by NiloCK 2y ago
To clarify: I didn't 'fix' the problem by re-inserting the package.json. I just gave a reminder that it that it was already in context. "I could confirm this root cause if I could see the contents of package.json".
270.
by NiloCK 2y ago
I learned this very explicitly recently. I've had some success with project and branch prompts - feeding a bunch of context into the beginning of each dialog. In one dialog, some 30k tokens later, Claude requested the contents of packa

More ›