lukev

17 ms·

331.
by lukev 2y ago
I have a thesis that RDF is good for this sort of work (I gave a talk on it at Clojure Conj last year, https://youtu.be/OxzUjpihIH4 if you're interested.) It's good for bridging the gaps between LLMs, ontologies,
332.
by lukev 2y ago
The OP's whole point is that he has money, not ideas. Plus, in your example, you could literally do both. Plus, money is not the same thing as IP or creativity. Plus, the post you were responding to was literally advocating giving the
333.
by lukev 2y ago
Sounds convincing! ...unless there were actually several thousand people who performed this experiment, got a negative result, and therefore don't remember it or post anything about it.
334.
by lukev 2y ago
This is pure capitalist brainworms. And I say this as someone who's not even particularly anti-capitalist.
335.
by lukev 2y ago
I know. It was a rhetorical question, the suggestion being that optimizing for pageviews might not actually be what creates the most value.
336.
by lukev 2y ago
On what measure of success? A/B tests say that Ariana Grande is about 20x better than Mozart (according to Spotify stats, actually impressed Mozart holds up so well.) So something besides popularity in the moment has to matter. Which A
337.
by lukev 2y ago
I'm not sure that's true. A LLM can code because it is trained on existing code. Empirically, LLMs work best at coding when doing completely "routine" coding tasks: CRUD apps, React components, etc. Because there's
338.
by lukev 2y ago
Code is a liability, not an asset. It is a necessary evil to create functional software. Senior devs know this, and factor code down to the minimum necessary. Junior devs and LLMs think that writing code is the point and will generate lots
339.
by lukev 2y ago
Most enterprise use cases also have strong authz requirements. You can't really maintain authz while fine tuning (unless you do a separate fine-tune for each permission set.) So RAG is the way to go, there.
340.
by lukev 2y ago
Except that the direction of art is usually influenced by the artists, without too much calibration to what the uninformed public wants. It’s actually a really interesting question if this affects the market, at all. Because if AI can gener
341.
by lukev 2y ago
Yeah but even then they won't describe it using the same sort of language that everyone else developing these things does. How many parameters? What kind of corpus was it trained on? MoE, single model, or something else? Will the weigh
342.
by lukev 2y ago
That may be how most people see art, but it is not how artists see art. Neither side is "right" or has a monopoly on the definition of art. You can't tell someone else how to think or feel about something.
343.
by lukev 2y ago
This is a digression, but I really wish Amazon would be more normal in their product descriptions. Amazon is rapidly developing its own jargon such that you need to understand how Amazon talks about things (and its existing product lineup)
344.
by lukev 2y ago
The use of non-experts as judges renders this meaningless. Imagine an algorithm that drew random shapes. Then ask people who've never engaged with art theory or history to compare its output with what they see in a modern art museum. Y
345.
by lukev 2y ago
Really? My partner and I and some of our friends thought the opposite. Skarsgård can be tough but in many of his roles he exhibits a fairly unconventional form of masculinity. My expectation is that he's going to be able to play an u
346.
by lukev 2y ago
Yes, and you wouldn't even need to do it in realtime as a user walks around. Generate incrementally using a pathfinding system for a bot to move around and "create the world" as it goes, as if a Google street view car followe
347.
by lukev 2y ago
[flagged]
348.
by lukev 2y ago
This is a good article on the complexities and interaction between ActivityPub servers, their existing ToS, and Bluesky. But as far as Bluesky itself is concerned, the answer is wonderfully simple: it's just public. Full stop. Posting
349.
by lukev 2y ago
Actually I think even this makes it look scarier than it is if you're on an M-series Apple. In that case, the steps to building llama.cpp are: 1. Clone the repo. 2. Run `make`. To start chatting with a model all you need is to: 1. Down
350.
by lukev 2y ago
Google has clearly transitioned away from prioritizing customer value (and content creator value, unless you're an advertiser) in favor of some internal opaque KPIs. After switching to DuckDuckGo years ago, and Kagi last year, it'
351.
by lukev 2y ago
I don't know what it's true to suspect, since clearly a lot of very smart people are working in the field, in places. It is empirically true that none of the industry discourse around leaderboards and benchmarks uses any of the te
352.
by lukev 2y ago
But we don't know if AI development is following an exponential or sigmoid curve (actually we do kind of, now, but that's beside the point for this post.) A wise institution will make decisions based on current capabilities, not
353.
by lukev 2y ago
I wasn't scientific about it, unfortunately. My searches were natural language, not token-based, though.
354.
by lukev 2y ago
I ran a couple needle-in-a-haystack type queries with just a 32k context length, and was very much not impressed. It often failed to find facts buried in the middle of the prompt, that were stated almost identically to the question being as
355.
by lukev 2y ago
Even taking this as completely true, what does it mean for the field? How can one become a professional, experienced engineer if one has not put in multiple years being a productive junior engineer? And what happens if those jobs are no lon
356.
by lukev 2y ago
I don't necessarily believe this for a second but I'm going to suggest it because I'm feeling spicy. OpenAI clearly downgrades some of their APIs from their maximal theoretic capability, for the purposes of response time&#x2F
357.
by lukev 2y ago
Good news for you: https://membership.theonion.com
358.
by lukev 2y ago
No no that's my point: Bluesky has no algorithm. You see only the posts of people you follow, in order, and nothing else (unless you deliberately subscribe to "discovery" feeds.) So you have complete control.
359.
by lukev 2y ago
If you are good about curating your follows, you can get a nice nonpolitical feed on Bluesky. That said this is definitely not the week to try to calibrate that, since everyone currently has Big Feelings even if they're normally not ov
360.
by lukev 2y ago
Not really, without explaining how the whole thing works. There's a lot of resources... my favorite is https://karpathy.ai/zero-to-hero.html

More ›