noosphr

39 ms·

961.
by noosphr 2y ago
For a blog post of 1,200 words the bitter lesson has done more damage to AI research and funding than blowing up a nuclear bomb at neurips would. Every time I try to write a reasonable blog post about why it's wrong it blows up to tens
962.
by noosphr 2y ago
Hot take: Vibe coding is going to be the new Excel of technical debt. Most tech savvy places will avoid it, most good programmers will never encounter it. A bunch or us will make a career out of fixing the mess it makes after it explodes. M
963.
by noosphr 2y ago
After throwing the average meeting as an mp3 to your system, yes, you have diarization solved much better than everyone else I've tried by far. I'd say you're 95% of the way to being good enough for becoming the backbone of m
964.
by noosphr 2y ago
Having read the docs - used chat gpt to summarize them - there is no mention of speaker diarization for these models. This is a _very_ low hanging fruit anyone with a couple of dgx h100 servers can solve in a month and is a real world probl
965.
by noosphr 2y ago
You don't need an authorization wall to have your stuff behind, you can just as easily use an anonymous micropayment service for each request. That we live in an internet where getting too many visitors is an existential crisis for web
966.
by noosphr 2y ago
Here we get to the original sin of packet networking. The ARPANET was never meant to be commercial or private. All the protocols developed for it were meant to be subsidized by universities, the government or the military, with the names, p
967.
by noosphr 2y ago
If you live in the new world this seems cool and interesting. If you live in these places it's annoying because it means every time you try and build something an army of archeologists descends on your property and stops you from doing
968.
by noosphr 2y ago
The jump from llama2 to llama3 had something to do with meta downloading every textbook ever published and using it as training data. The arguments by meta so far in that court case are absolutely terrible and I'm half expecting to see
969.
by noosphr 2y ago
>As a leader and teacher with more grey hair every day, I feel this responsibility You should probably do less of that since: 1). That's not what any of my posts said. 2). My degree says mathematical physics and not computer science
970.
by noosphr 2y ago
Not sure where you're getting that from my post. The point is that all other engineering disciplines don't have enough design space to make a true mess of their projects. Software by comparison hasn't been limited in its desi
971.
by noosphr 2y ago
Competition can only work when there is variation between the entities competing. In the US right now you can have a death match between every AI lab, then give all the resources to the one which wins and you'd still have largely the s
972.
by noosphr 2y ago
>However, at the end of the day, there is an input and output and compute and memory needed to run the thing and if we look at that we realize, we never actually left the bounded physical realm and we can still engineer software systems
973.
by noosphr 2y ago
Software is different. All other engineering disciplines are ultimately limited to building things in (at most) 3 euclidean dimensions. There is only so much junk you can hide in a finite volume of space. Code by comparison lives in hyperbo
974.
by noosphr 2y ago
It's not just them. Everyone is scrambling. US tech, and western tech in general, is very culturally - and by this I mean in the type of coding people have done - homogeneous. The deep seek papers published over the last two weeks are
975.
by noosphr 2y ago
That's fair enough, but you should test other frontier model types to see if the benchmark makes sense for them. For example the shortest path benchmark is largely useless when you look at reasoning models - since they have the equival
976.
by noosphr 2y ago
>We evaluate six frontier language models across both settings: Claude 3.5-Sonnet, GPT-4o, GPT-4o-Mini, Deepseek-v3, Gemini-2-Flash, and Llama-3.3-70B-Instruct. While I appreciate the effort and creativity that went into this there are a
977.
by noosphr 2y ago
Any sufficiently advanced plotting library with an api that can be called externally becomes indistinguishable from a GUI toolkit: https://www.gnu.org/software/guile/docs/guile-tut/tutorial.h... Not sure
978.
by noosphr 2y ago
Then why on gods green Earth do you also have: > HIRING CHALLENGE: Find our recenter browser plugin (it is open-source). Understand what it is for and suggest an interesting and personally useful feature you can add to it. Go through the
979.
by noosphr 2y ago
How about you ask for a personal repo with open source projects that solve real problems? Instead of inventing a bullshit pre-job you can make the world a better place. You also get a strong signal on who can deal with real world problems w
980.
by noosphr 2y ago
This was done as part of a contract and like I said I never got permission to write a paper about it, let alone share code.
981.
by noosphr 2y ago
It was a SAT solver with the symbolic expression fed into the LLM. Using the SAT solver I'd rate how well the LLM was able to solve a given boolean formula, if it got it wrong it would get a second prompt to ask it to split the boolean
982.
by noosphr 2y ago
Above sota on logical reasoning for grounding text. LLMs at the time were so bad at it that even fornier models when given A, not B would derive A and B in the output about half the time.
983.
by noosphr 2y ago
This is secret sauce that people have been hoarding for the last year or so. With the deepseek open source releases this is now worth a lot less and companies are cashing out on reputational increases instead of being scooped. I have done t
984.
by noosphr 2y ago
After four years of "This VLM will solve OCR forever!" posts I've firmly put VLMs in the "useless until 100t parameters" category. Some day when there is enough internal state and training data that they can recogni
985.
by noosphr 2y ago
>And boards are largely made up of senior management of other companies, scratching each other's backs. My experience with boards is that everyone on them would rather be doing something else. If they could offload 90% of their work
986.
by noosphr 2y ago
All the names and places have been changed to protect the guilty: When I was a young data engineer in 2014 at a very large telco in Canada the real time resource allocation of network bandwidth would work flawlessly from 6am to 11pm, but wo
987.
by noosphr 2y ago
The ultimate authority in any public company is the board. The majority of employees will never have to think about it, but everyone from the CEO down is ultimately hired at their discretion. To see how things normally operate read up on th