llm_trw

11 ms·

91.
by llm_trw 2y ago
Never hire zealots who don't share your religion, they may be cheap - or free in this case - but in the long run they cost you a lot more.
92.
by llm_trw 2y ago
I resent that. There are a few non-linear function operations in between the matrix multiplications.
93.
by llm_trw 2y ago
Twitch plays thermonuclear war.
94.
by llm_trw 2y ago
Leading in AI safety is leading in AI lobotomization.
95.
by llm_trw 2y ago
Ironically enough I have a script that strips all non-ascii characters from my screen reader because I found that _all_ online text was polluted with invisible and annoying to listen to characters.
96.
by llm_trw 2y ago
They had some sort of revolution the previous few centuries too. Pretending that Europe wasn't in a perpetual blood bath since the end of the Pax Romana until 1815 shows a gross ignorance of basic facts. The printing press was a net po
97.
by llm_trw 2y ago
All regulations are political. I'm not sure what you think the issue here is.
98.
by llm_trw 2y ago
The simple solution is that all such devices must be submitted to a central government database and all blue prints, source code etc. will be released into the public domain is the company no longer supports the device or closes down.
99.
by llm_trw 2y ago
That code is more easily readable than most visual basic business logic running billion dollars in trades I saw in the 00s.
100.
by llm_trw 2y ago
The summer of 9/11 I brought on a full suit of hoplite armor with shield and spear on board a regional plane. The stewardess told me I can't have it in the passenger compartment so I should just give it to the pilot before takeoff
101.
by llm_trw 2y ago
If you're interested please drop an email. I've only worked deeply with pipelines extracting data from documents and I'd be interested in hearing what the challenges with databases are.
102.
by llm_trw 2y ago
>So let me get this straight- we are going to train AI models to perform screen recognition of some kind (so it can ascertain layout and detect the "important" ui elements), and additionally ask that AI to OCR all text on the s
103.
by llm_trw 2y ago
>Digging into use cases you’d fine that for a particular question you needed to not just get all the rows from a column, you needed to do some obscure JOIN ON operation. This fact was only known by 2 data scientists in charge of writing
104.
by llm_trw 2y ago
The intern under the analyst did.
105.
by llm_trw 2y ago
> we only have months to years of experience with the mistakes that LLM's and other AI's make. The mistakes are also very much model dependent. That you have build a system which improves the accuracy of one models output give
106.
by llm_trw 2y ago
>You just ruled out 99% of humans from having reasoning capabilities. After a conversation with humans I think you'd agree 1% of them being able to reason deeply is a vast overestimation. A good example to see how little people can
107.
by llm_trw 2y ago
I don't get offended when people call my work a stochastic parrot. I just put them in the same bucket of intelligence as an 8b model and weight their inputs accordingly.
108.
by llm_trw 2y ago
>goes through personal experience of the evaluator Real reasoning is being able to manipulate symbolic expressions in a consistent manner while preserving some invariants. Personal experience as logic is how you end up with the Holocaust
109.
by llm_trw 2y ago
I'm happy to explain my strategies about synthetic data - it's just that you'll need to hear about the onions I wore in my day: https://www.youtube.com/watch?v=yujF8AumiQo
110.
by llm_trw 2y ago
I've literally build a dynamic bench mark where I test reasoning models on their performance on deriving conclusions from assumptions through sequent calculus. o3-mini high effort can derive chains that are 8 inference rules deep with
111.
by llm_trw 2y ago
I thought I'd give this the benefit of the doubt. It's trying its hardest to be put in the "this is bullshit" pile. Five lines of content spread through five pages through the magic of parallax scrolling. Examples wander
112.
by llm_trw 2y ago
I'd strongly suggest having a look at the Cardiac [0] and MIX [1] computers. Shift operations are pretty important and having a dedicated multiplication/division is also very useful. The fact that all of these are (or can be) deci
113.
by llm_trw 2y ago
So build a kernel that uses the same API as Linux but in Rust. Surely it should be easy if you're asking the Linux people to do the same thing for drivers written in Rust?
114.
by llm_trw 2y ago
They don't. R1 gets the right answer in the thinking part of the response and ignores it in the response more than half the time in my tests.
115.
by llm_trw 2y ago
If you don't want to use C perhaps you shouldn't contribute to a 33 year old project written in C?
116.
by llm_trw 2y ago
You can generate the data synthetically. We never had the budget to do it but I do have some notes somewhere on a 2d context free grammar to generate arbitrarily nested rows/columns and a css styling that got applied to the xhtml outpu
117.
by llm_trw 2y ago
This is a response to: https://news.ycombinator.com/item?id=42952605 A fun threat to read for the current hype cycle. You can tell who is working in the field by the fact they don't use VLMs for OCR and who isn't
118.
by llm_trw 2y ago
It's not possible with current gen models. To even have a chance at doing it you'd need to start the training from scratch with _huge_ penalties for filling in missing information and a _much_ larger vision component to the model.
119.
by llm_trw 2y ago
This is great until it hallucinates rows in a report with company assets that don't exist - why wouldn't a mining company own some excavation equipment? - and pollutes all future searches with fake data right from the start. I lau
120.
by llm_trw 2y ago
They are trying their best to lose the current generation to no one. It's pretty funny when your own engineers can't get access to the cards you've supposedly launched.

More ›