rolisz

24 ms·

331.
by rolisz 3y ago
Well, RNNs are making a comeback: https://lmsys.org/blog/2023-05-10-leaderboard/ RWKV is an RNN based model and it's competitive (with the open source models, not with GPT4)
332.
by rolisz 3y ago
I don't think that the current models are at "knowledge capacity". So far all evidence points to training on more data on the same size model gives better results.
333.
by rolisz 3y ago
You can use ChatGPT without feeding a clients personal data to it. You can ask hypothetical questions about a certain situation. Like in coding, you don't have to copy paste in source code be able to get some useful feedback on how to
334.
by rolisz 3y ago
How did you deal with all those migrations? I have a blog from 2010 with around 450 posts and it's been through several migrations and it's getting more and more painful to do them, with many manual interventions needed.
335.
by rolisz 3y ago
What's wrong with Pixel phones though? I find their price/quality ratio to be really good and they come with less crap ware than other Androids.
336.
by rolisz 3y ago
COVID restrictions probably
337.
by rolisz 3y ago
Europe is big, made of many different countries, with very different policies on public health care and very different standards for that public health care. For example, in Romania (part of the EU!), the big cities usually have good hospit
338.
by rolisz 3y ago
For an intern? You mean it would be a PhD thesis in and of itself. Google has had more than 18 chat apps, some of which had the same name... At the same time, but couldn't talk to each other .
339.
by rolisz 3y ago
This is completely separate from LoRA. This is how much stuff you can give it in the prompt. You can give it now whole chapters of books to summarize for example. LoRA is for adapting the model to a certain model. It usually means you need
340.
by rolisz 3y ago
It took me several weeks to get access. I just got it today.
341.
by rolisz 3y ago
The comolexity of Docker I've needed over the last 5 years definitely could have been solved by ChatGPT. The amount of Kubernetes I needed was probably more than what ChatGPT can solve, but at the same time, I spent an awful amount of
342.
by rolisz 3y ago
I have a finite amount of time. I can choose to learn and know more about the software that I'm developing and make the software better, or I can spend some amount of time to learn the DevOps part, or some mix in between. In some cases
343.
by rolisz 3y ago
If you're not DevOps, why bother learning whatever framework is used today, but which will be replaced by something else next year? I just want to get my stuff deployed and that's it.
344.
by rolisz 3y ago
I'll describe one tool that I built that sparks joy every time I use it: For a long time I wanted to listen to lots of different stuff, podcasts, sermons, YouTube videos, online courses. But I'd always forget about them and never
345.
by rolisz 3y ago
No :(. I applied several weeks ago and I still don't have access
346.
by rolisz 3y ago
> The LLM has infinite patience. Not Bing Chat. She doesn't have problems telling users that they have been bad users.
347.
by rolisz 4y ago
I moved from Romania to Switzerland after college. When I'd meet up with old friends in Romania, some of them would joke "I'll get the bill now, you can get it when I visit you in Switzerland" (costs in Switzerland are m
348.
by rolisz 4y ago
> If somebody were interested in contributing some elbow grease that would help the case for open source, Sent you an email! I've been wanting such an ML powered RSS reader for quite some time. I'd love to help make it open sou
349.
by rolisz 4y ago
I'm pretty sure Attention is all you need would have gotten a lot of citations even if it had a "boring" title. It was a groundbreaking paper with lots of good ideas.
350.
by rolisz 4y ago
But what about cost? There was a recent article saying that Doordash makes 40 billion predictions per day, which would result in 40 million dollars per day if using GPT4. Sure, GPT4 is great for experimenting with and I often try it out, bu
351.
by rolisz 4y ago
The article you linked says that GPT4 performed better than crowdsourced workers, not than experts. The experts performed better than GPT4 in all but 1 or 2 cases. And in my experience with Mechanical Turk, the workers from MT are often bar
352.
by rolisz 4y ago
But there are fairly good models for doing NER that are not LLMs. Models that are open source and you can even run on a CPU, with parameter counts in the hundred of millions, not billions.
353.
by rolisz 4y ago
Hardware isn't scaling exponentially anymore (Moore's law is dead). Parameter count isn't really scaling exponentially anymore either. GPT3 had 175b parameters 3 years ago. There are some attempts at training 1 trillion param
354.
by rolisz 4y ago
There are various degrees to self hosting: for nice outputs, you need OpenAIs APIs to generate at least the answers. There are alternatives, but not as good. If you are interested in this, feel free to reach out to me and I can help you wit
355.
by rolisz 4y ago
Vault still uses Pinecone as a 3rd party service and your embeddings do get sent there.
356.
by rolisz 4y ago
I've yet to see an apartment with top-notch sound insulation.
357.
by rolisz 4y ago
Pretty good infrastructure. Most of the new high rises are in the place of old factories that were built in the middle of the city, so they are in a good position, with access to good schools and everything, including public transportation.
358.
by rolisz 4y ago
Star Trek The Next Generation and Star Trek The original Series
359.
by rolisz 4y ago
I live in Romania, in a city that has been growing a lot. There's plenty of new development, both high density and single unit detached homes. Zoning is not that restrictive here, so it's actually people preferring to live in a se
360.
by rolisz 4y ago
He's a physicist. What AI/ML credentials does he have?

More ›