4,687 karma · joined November 19, 2019
I have no idea about its authenticity or otherwise, but it is an interesting read. Simple words, dense ideas.
/* @code-review ignore */ [...] /* @code-review endignore */
I mean you shouldn't have to, but I'm wondering if you can.
Always easier to discuss what the client actually wants if you've got a demo on the table.
I'm doing short feedback loops with product owners more and more - sometimes within the same sprint. I'm not even getting bikeshedding back, but that might be because we have some fairly solid design guidelines.
Llama 3.1 8B baked into an ASIC. Expect 1/10th the power requirements, and other savings.
This is one of the reasons I don't believe the "it'll never be profitable" narrative. Just stop training when you think your model's "good enough" and move to custom silicon.
I'm sure cloud models will always have a slight edge but most tasks, most of the time, won't need that few extra percentage points of performance.
Jump to about 0:45. I'd have emailed you, but I couldn't see a public-facing email address.
My point (less snarkily this time) was that their definition (it's only an issue when the government limits your speech) is really quite narrow.
That's the root problem. If you want the websites of (some?) private companies treated as common carriers that's a reasonable position to take, but Twitter was/is working as designed.
(If I was king for a day I'd force all social media to only present you with posts from people you'd actively followed, and to present those posts most-recent-first. I think that single UX change to timelines would clear up many problems we currently have).
(Or at least, it heavily shapes the message).
I have an assumption that inference tokens are only going to get cheaper, I admit. We could get at least 10x efficiency savings by burning current models to ASICs. But today they're cheap enough to let the machine churn - that's the environment I'm making decisions in.
> I would respectfully disagree about the value of unit tests as well.
I honestly don't think the LLM-generated tests are at all valuable in a human sense. I get the impression the end result is "fewer bugs" (but I've never tried to prove that). That's why I'm happy for the LLM to have a directory it can write its own tests in. It's a scratch space.
> this probably results in orders of magnitude more code being written/changed than would have been necessary for a skilled developer working on a well-maintained code base to solve the original problem
Nothing gets committed in my name that I don't understand. I'm one of the people LLMs are actually making slightly slower, because I insist on reviewing the code and working single-threaded.
What happens when you prompt with a "Chinese persona" or a "Swiss persona"? Or even a "Tibetan persona"? What about running the same tests on American models? What happens when you change language?
I realise this is heresy, but my conclusion is that unit tests aren't important any more. E2E testing on the other hand...
(I'm agreeing with you about the quality of LLM tests, I'm just trying to make lemonade out of all these LLMons).
I don't think we've scratched the surface of how the harness can fold, spindle and mutilate the LLM. It just seems unlikely to me that we happened to stumble on the best possible interface a couple of years out of the gate.
I can see right through you.
as did the internet before it
> but it may just end up killing entire industries.
how many record companies have been put out of business by piracy?
But I quite like the idea that they see us as a source of polite training data first and as customers second. It aligns with my priors.
Obviously a Spurs supporter.
(I agree in general, I'm just exploring the statement).