npip99

13 ms·

1.
by npip99 29d ago
just add to markdown instructions to check that this failure mode isn't happening in its diff before yielding back to human review. Any issue that occurs more than once, you can just put in the review markdown instructions so you don&#
2.
Show HN: AutoGPU – AI designs a real 7nm GPU, from Verilog to GDSII(github.com)
4 points by npip99 3mo ago|0 comments
3.
by npip99 1y ago
Bradley-Terry and Elo scores are equivalent mathematical models! The fundamental presumption is the same Thurstone model - that an individual's skill in a particular game is a normally distributed random variable around their fundament
4.
by npip99 1y ago
Yeah absolutely. In your link, it iterates on _ = ^{_}, until it finds the fixed point. In our training pipeline, we had to convert the fixed point iteration to be on _ directly for numerical stability. I have a post on that here!: https:&
5.
by npip99 1y ago
In our case training and inferencing the models takes days, calculating all of the ELOs take 1min haha. So we didn't need to optimize the calculation. But, we did need to work on numeric stability! I have our calculations here: - ht
6.
by npip99 1y ago
We tried a bradley-terry loss function, as calculated with https://hackmd.io/@-Gjw1zWMSH6lMPRlziQFEw/SJ8sRl1Zge We found that MSE after elo-adjustment worked equally well. And, MSE lets you shuffle (q, d) across the da
7.
by npip99 1y ago
Yes our pairwise method is based entirely on 2AFC comparisons, for both intra-query and inter-query ELO calculations. It's definitely the best if not only way to get extremely high signal, and a score assignment that actually converges
8.
by npip99 1y ago
Yeah that's exactly what we observed. Our goal was to create an absolute score that's completely independent from the Corpus, which is difficult because naturally all ELO distributions are inherently tied to the corpus itself! Whe
9.
by npip99 1y ago
I often see it rendered as "Elo" but I've always found it more natural to capitalize as "ELO", but perhaps I should swap to "Elo" given this. Pronouncing "ee-low" is certainly the way it's d
10.
by npip99 1y ago
Hey! We actually did a lot of research into ELO consistency, i.e. to check whether or not the NxN pairwise matrix followed the ELO model. It was a long road that's probably grounds for an entirely separate blog post, but the TLDR is th
11.
by npip99 2y ago
Yes, a step where you do a structured extraction into a database column would be a potential solution. But, it requires a preprocessing step. It all depends on the use-case, sometimes you get a query that you couldn't have predicted th
12.
I looked at 1000s of RAG queries to figure out the problem with semantic search
6 points by npip99 2y ago|3 comments
13.
by npip99 2y ago
Good question! For long documents we have a rolling window strategy. So, we cut the document into 5,000 token groupings for use in inference. There's also a 400 token overlap, and we prefer the earlier chunk for overlap tokens. For exa
14.
by npip99 2y ago
Yep benchmarks are available at https://github.com/ZeroEntropy-AI/llama-chunk?tab=readme-ov-... , we used this dataset https://github.com/ZeroEntropy-AI/legalbenchrag which is a retrieval-focused
15.
LlamaChunk: Better RAG Chunking Than LlamaIndex(github.com)
15 points by npip99 2y ago|5 comments
16.
by npip99 2y ago
Hey HN! I've written a lot of RAG pipelines over the last year, and one consistent pain-point is writing regex to chunk the documents correctly. Right now, the most common chunking algorithms are: - Split every 1000 characters - Split
17.
by npip99 5y ago
"Who wouldn’t prefer to stay home if they had the choice? I could spend my days traveling" It's not travelling... If you're staying at home...
18.
by npip99 5y ago
That's because the $50M were owned by a large percentage of the users of Ethereum, not a single person. It represented 15% of the total ETH in circulation, back when the currency was in a very nascent state and the flagship product tha
19.
by npip99 6y ago
Wait a second, am I missing out right now? I have an opportunity at Jane Street, but I haven't been exposed to any of the other small name firms. Where do I look to find the higher paying jobs?
20.
by npip99 7y ago
This seems like a waste of time. It's not illegal to register a domain name, no matter how similar it is to your domain name. If you want those domain names, then buy them. My family runs a business and we bought the domain names that
21.
by npip99 7y ago
I mean, I can be more lenient. Europe had good researchers. They were the Athens, between the Germans ones of the 20-50's, and Turing, there was a lot. The Russians got a lot done too. But again, all research. Nothing profitable. It wa
22.
by npip99 7y ago
Eh, I don't think so. There's too much interdependence, you can have anxiety over how many things your dependent on, especially in Tech. It's actually terrifying. If Azure closed down tomorrow I'd be fucked so hard, thou
23.
by npip99 7y ago
I'm not sure what you mean by "Quality of life is just a sham invented by politicians to keep the plebs from demanding... quality of life?". I don't think politicians invented anything like that? I remain confused.
24.
by npip99 7y ago
This is kind of what I touch on at the end, with the getting the kids into business, there is a moral concern there. The first executives are intelligent, and caring. They're the engineers. Once you have MBAs flooding in, which HR does
25.
by npip99 7y ago
I mean if you have $100k in wealth you absolutely have travel freedom. Backpacking in Europe is a common experience so it's pretty cheap and the rest of the countries have crazy USD conversion rates. $100k could be a lifetime's wo
26.
by npip99 7y ago
The article says "Emits more pollution than all of Europe's Cars", which is a bit of a garbage claim, because obviously all of the CO2 is coming from the cars.
27.
by npip99 7y ago
Why is giving a "Delivery Unlimited" option considered to by "Taking on Target Shipt, Amazon Prime, etc"? It's just a reorganization of what you pay, since you're going to pay for those shipping costs one way o
28.
by npip99 7y ago
As much as this technology is powerful, you can't make it "illegal" to take photos of a public park, and then bring them home and do whatever you want with them at home. That would just be absurd. Banning the software online
29.
by npip99 8y ago
None of the top comments make any sense to me. I simply don't understand them. Anyway, here is my understanding of the subject: (1) Yes, Uber should have trained the driver better to look at the road. By trained, I mean there should ha