HNHacker News
TopNewBestAskShowJobs

sjwhevvvvvsj

448 karma · joined January 2, 2024

submissionscomments
sjwhevvvvvsj··on Hi everyone yes, I left OpenAI yesterday
It’s not fixed whatsoever. Mistral 7B runs on a MacBook Air, and it’s free. Zero cost LLM, no network latency.
sjwhevvvvvsj··on Hi everyone yes, I left OpenAI yesterday
Yes but if a 7b LLM will give you the same “Hello World” as the 70b, and that’s literally all you need, using a bigger model is just burning energy for no reason at all.
sjwhevvvvvsj··on Hi everyone yes, I left OpenAI yesterday
So I think that’s a “your problem isn’t right for the tool” issue, not a “Mistral isn’t capable” issue.
sjwhevvvvvsj··on Hi everyone yes, I left OpenAI yesterday
Yes, but for certain classes of problems small LLM are highly performant - in many cases equal to a GPT-4, which sure can do more things well, but adding 2+2 is gonna be 4 no matter what. You don’t need a tank to drive to the grocery store, just a small car with a trunk.

So the assertion that small models aren’t as good just isn’t correct. They are amazing at certain things, and are incredibly faster and cheaper than larger models.

sjwhevvvvvsj··on Hi everyone yes, I left OpenAI yesterday
People think LLM are all or nothing, like it’s either god-like AGI or it’s useless “hallucinating”.

In reality you have to know the strengths and weaknesses of any tool, and small/fast LLM can do a tremendous amount within a fixed scope. The people at Mistral get this.

sjwhevvvvvsj··on Mozilla downsizes as it refocuses on Firefox and AI
Yup, and that’s the thing with the LLM craze - it’s super energy intensive and for the most difficult problems you still need really good hand curated data anyway.

My point is more “we know how to do this already and it’s just a hard problem”.

sjwhevvvvvsj··on Hi everyone yes, I left OpenAI yesterday
Who says it’s wrong? I have very discrete tasks which involve resolving linguistic ambiguity and they can perform very well.
sjwhevvvvvsj··on Mozilla downsizes as it refocuses on Firefox and AI
The loudest people have the least to say, but I think there’s a lot of smart use cases beyond variation on chat bots.
sjwhevvvvvsj··on Mozilla downsizes as it refocuses on Firefox and AI
#2 is just a document classification task, you can do it with DistilBERT.
sjwhevvvvvsj··on Mozilla downsizes as it refocuses on Firefox and AI
Hey now, Google paid very good money for that antitrust testimony.
sjwhevvvvvsj··on Mozilla downsizes as it refocuses on Firefox and AI
Usually somebody serving in an interim position doesn’t make big strategic changes.
sjwhevvvvvsj··on Hi everyone yes, I left OpenAI yesterday
What Mistral has though is speed, and with speed comes scale.
sjwhevvvvvsj··on Hi everyone yes, I left OpenAI yesterday
He’s first author on a ton of those papers. That’s a tenure worthy CV almost anywhere. Gimme a break.
sjwhevvvvvsj··on Nvidia is now more valuable than Amazon and Google
the answer is it’s a state sanctioned monopoly which concentrates control of advanced technological processes in the hands of NATO.

ASML is a strategic asset.

sjwhevvvvvsj··on Nvidia is now more valuable than Amazon and Google
If AMDs version of “asleep at the wheel” is them eating Intels lunch for the last few years, I’d hate to see them awake.

Intel is weak, and it’s much smarter to put resources into squeezing Intel while they are on the ropes than shift too much focus to ML.

sjwhevvvvvsj··on Orbital's Hartnoll brothers look back
http://hyperreal.org/

:-)

sjwhevvvvvsj··on GeneGPT, a tool-augmented LLM for bioinformatics
I was hoping for a Mistral fine tune at the least.
sjwhevvvvvsj··on Developers revolt over Apple's new app store rules
This is a great way to encourage developers to build the app that will show an essential use case for the Vision Pro.

It’s not like Vision Pro is a solution in need of a problem, desperately waiting for a killer app, or that apps built the iPhone ecosystem into the juggernaut it is now.

Apple doesn’t need developers, right? Right?

sjwhevvvvvsj··on Orbit: C/C++ Performance Profiler
Did anything come of that?
sjwhevvvvvsj··on Orbit: C/C++ Performance Profiler
Stadia was such a monumental technical achievement, it was painful to see how they killed it without even seeing what it could be used for. Idiocy.
sjwhevvvvvsj··on Orbit: C/C++ Performance Profiler
Looks like it may have been stadia related.
sjwhevvvvvsj··on What it was like working for Gitlab
Also the social safety net in NL covers a lot of things you have to pay for yourself in the USA.
sjwhevvvvvsj··on Niagara Launcher
Yeah, and equal time spent futzing with your launcher is better spent tinkering with a raspberry pi or similar. At least then you may develop a skill.

Android customization is sort of like junk food - enjoyable in the moment, but not meaningfully improving your life.

sjwhevvvvvsj··on Bard is now Gemini, and we’re rolling out a mobile app and Gemini Advanced
It’s not just media narrative. The culture was eroding for years, as you note, but the dam finally broke and they went full IBM/Kodak. Or in other words, “slowly at first, then all at once”.
sjwhevvvvvsj··on Bard is now Gemini, and we’re rolling out a mobile app and Gemini Advanced
I think Google lost the top researchers when they destroyed the culture. All the competitor companies are mainly led by ex-Google talent, and honestly who in their right mind would take a Google job today over OpenAI, Mistral, or even Meta (where you will be releasing models like Llama for the world to use).

Google killed the culture and is bleeding top talent. They have reduced themselves to digital landlord and sure they can extract rent, but that’s not what attracts people.

sjwhevvvvvsj··on Automate the creation of YouTube Shorts by providing a topic
“Your work”, is their “content”.
sjwhevvvvvsj··on Automate the creation of YouTube Shorts by providing a topic
And then you can double dip with a little drop shipping on the side.
sjwhevvvvvsj··on Automate the creation of YouTube Shorts by providing a topic
Don’t hate the player, hate the game.
sjwhevvvvvsj··on Automate the creation of YouTube Shorts by providing a topic
Ruth approves.
sjwhevvvvvsj··on Ask HN: Best "entry tech" jobs to learn coding?
In all honesty, at the rate LLM coding ability is increasing I am not sure how many entry level roles will be left for people who don’t have a solid grounding in computer science.

Generally speaking, I believe this is a bad thing as coding careers have historically been open to anybody with the aptitude.

I’m not sure anybody can provide solid advice beyond “don’t give money to a boot camp”.

← PreviousPage 2 of 5Next →