p1esk

75 ms·

961.
by p1esk 3y ago
Even if so, the next gen model will fix it.
962.
by p1esk 3y ago
How to create my own LLM? Step 1: get a billion dollars. That’s your main trade secret.
963.
by p1esk 3y ago
I’m both excited and scared to think about this “significant improvement” over GPT-4. It can make our jobs a lot easier or it can take our jobs.
964.
by p1esk 3y ago
How do these small models compare to gpt4 for coding and technical questions? I noticed that gpt3.5 is practically useless to me (either wrong or too generic), while gpt4 provides a decent answer 80% of the time.
965.
by p1esk 3y ago
the handymen, plumbers, and electricians of the world will be mostly unaffected. Their work is hard to automate. It’s funny, when I started in ML 10 years ago people said the same thing about writers, artists, musicians. And yet here we a
966.
by p1esk 3y ago
It’s better to think about mapping groups of weights to an entry in a large codebook, and then using clever methods to do weight lookups. 2 bits of precision per weight is perfectly fine as long as you have enough weights. The information e
967.
by p1esk 3y ago
This is for GPUs, not CPUs. GPUs do have lower precision ALUs to do math on fewer bits. Though not 2 bits - I believe there’s support for 1, 4 and 8 bit computation in modern Nvidia cards. But even without such support there’s a benefit
968.
by p1esk 3y ago
There are far more people who post obviously wrong, confusing and dangerous things online with total conviction. There are people who seriously believe Earth is flat, for example.
969.
by p1esk 3y ago
Not Ultra version
970.
by p1esk 3y ago
I haven’t read it, but looks interesting!
971.
by p1esk 3y ago
I don't see much difference between brain uploading and having children.
972.
by p1esk 3y ago
We don't know that. No one has demonstrated it. It's very likely that at larger scales, for a given amount of compute you cannot train a traditional RNN to be as a good as a transformer.
973.
by p1esk 3y ago
A quote from the link you posted: "Then, when you refer to “Lambda”, “ChatGPT”, “Bard”, or “Claude” then, it’s not the model weights that you are referring to. It’s the dataset." Yet all four of these examples use the same model a
974.
by p1esk 3y ago
It is pretty bad for classical composers - groups together completely unrelated, very different people: I typed Monteverdi and the closest neighbors shown are Scriabin, Britten, Poulenc, Faure, Schubert. It does not make any sense.
975.
by p1esk 3y ago
Artificial synapse, not neuron.
976.
by p1esk 3y ago
Wait, how could "50% more cores, and 75% more memory bandwidth" result in anything less than 50% of better compute performance?
977.
by p1esk 3y ago
Are there any surprising headlines today?
978.
by p1esk 3y ago
These two belief systems are compatible. Both groups believe everything was created by some higher being, who might be watching us right now.
979.
by p1esk 3y ago
Just to clarify, because this advise might be misleading. These LambdaLabs prices are pretty much meaningless, because there are no available instances currently, and haven't been for months. The last time I saw an available _hourly_ A
980.
by p1esk 3y ago
Half of the internet probably believes in God. Does that surprise you?
981.
Rethinking Brain Uploading Given 10-Year AI Timelines(youtube.com)
1 points by p1esk 3y ago|0 comments
982.
by p1esk 3y ago
Not clear how they determine “net worth” in countries like Russia. Many rich individuals there make money through corruption and the money are being hidden in various ways.
983.
by p1esk 3y ago
Your body is a result of a billion year old evolutionary optimization process. GPT-4 was trained from scratch in a few months.
984.
by p1esk 3y ago
Are you working on all that?
985.
by p1esk 3y ago
Gpt4 has 128k context length now.
986.
by p1esk 3y ago
LLMs might already be good enough to degrade the benefit of freedom of speech via signal-to-noise ratio Soon (1-2 years) LLMs will be good enough to improve the general SNR of the web. In fact I think GPT-4 might already be.
987.
by p1esk 3y ago
Well, if you haven’t tried gpt4 you haven’t really experienced modern AI.
988.
by p1esk 3y ago
Yes, it can be done and is done when needed, but your life will be simpler if you can fit your models in GPU memory. Even better if you can fit it in a single GPU memory.
989.
by p1esk 3y ago
I don’t get it - aren’t Jews considered white?
990.
by p1esk 3y ago
Google's advantage here is so blatantly obvious Exactly. Google has so much more resources, tries so hard to compete (it's literally life or death for them), and yet it's still so far behind. It's strange that you don&

More ›