jah242

20 ms·

1.
The $2T question: Will knowledge-work complexity scale with model capability?(twitter.com)
1 points by jah242 2mo ago|1 comments
2.
by jah242 2mo ago
I think an under-discussed aspect of the frontier-versus-commodity intelligence debate is that the rate at which real economic tasks adapt in complexity may be just as important as, if not more important than, the lag between open and close
3.
Knowledge factories and the industrialisation of knowledge work(sontoresearch.ai)
3 points by jah242 2mo ago|2 comments
4.
by jah242 2mo ago
I wrote this up while building Sonto. Essentially, I think the discussion around AI agents may be over-indexing on autonomous task horizons. METR-style task horizons clearly matter, but if the industrial revolution is anything to go by, red
5.
LLMs in Public Health – Part 2(joshuaharrissite.substack.com)
1 points by jah242 1y ago|0 comments
6.
LLMs in Public Health(joshuaharrissite.substack.com)
1 points by jah242 1y ago|0 comments
7.
Slow AI adoption – the arguments I find most compelling(substack.com)
1 points by jah242 1y ago|2 comments
8.
by jah242 1y ago
Interested to hear people's thoughts on why even though AI developments seem so revolutionary it may actually still not be the explosion of growth and societal change that often seems accepted in AI circles.
9.
by jah242 3y ago
This is text + image -> text but pretty cool and still might be of interest to you: https://llava-vl.github.io
10.
by jah242 3y ago
This is text + image -> text but pretty cool and still might be of interest to you: https://llava-vl.github.io
11.
by jah242 4y ago
This might be of interest to you (Google are getting there :))- https://palm-e.github.io
12.
by jah242 4y ago
Google actually recently went some steps further and combined the PaLM LLM (bigger than GPT-3.5) with a 22 billion parameter Vision Transformer to do this - https://palm-e.github.io
13.
by jah242 4y ago
Whilst maybe GPT-4 will change this, I think it is important to remember that these general ChatBots are not the way we have generally trained LLMs to write the best code. In fact, coding is one of the few areas where training specifically
14.
by jah242 4y ago
Whilst not the same I recommend you look at the DeepMind Gato paper to see surprisingly (relatively) simple multi modal can be - https://openreview.net/forum?id=1ikK0kHjvj Essentially to merge lots of modalities they just g
15.
by jah242 4y ago
I m sorry but in stating the goal posts haven't moved, you've literally just moved the goal posts. 'everything a human can do' is not the same as 'anything any human can do as well as the best humans at that thing (
16.
by jah242 4y ago
So I appreciate the jist of your point but the way these models work is rather more complicated that copying and pasting snippets and so it certainly is not 'literally' true. The models are trained to predict sub-word level tokens
17.
by jah242 4y ago
> but all it does is regurgitate existing works I am genuinely intrigued by this point of view and so would love to hear people who hold it's reasoning. Over the last few days I've seen hundreds of poems and stories from ones a
18.
by jah242 4y ago
Thanks! A less impressive but fun variant of this I am using is for open ended text adventure: > I want you to act as a text based adventure game. I will type commands and you will reply with what the adventure game should show. I want y
19.
by jah242 4y ago
For context, looking at UK data: 1. Yes incidence of cancer in 25-49 year olds has increased 22% from 1993 to 2018 - but that is 22% on a very low number which means it is still a very low number. When you account for increased screening, g
20.
by jah242 4y ago
> Secondly, the tropical hole wouldn’t have qualified as a hole under the old definition of a hole, which was defined as having ozone levels below a certain threshold. Lu’s study redefines a hole as an area with ozone loss greater than 2
21.
by jah242 4y ago
> correlated surge in the 45-64 bucket, the last two charts in the link I posted. Correlated surge is an absurd way to describe something that basically fluctuates around 0 over the course of a year, which makes me think this definitely
22.
by jah242 4y ago
Just in case you are asking in good faith. If you look at the same data release, the excess deaths are almost entirely in cancer and dementia in old people (with the exception of a spike in diabetes at the same time as the spike in COVID in
23.
by jah242 4y ago
I second this and can't recommend Keynes and his writing highly enough. I would also add that if you can get hold of Volume 12 of The Collected Writings of John Maynard Keynes it goes through his time as an investor and all related cor
24.
by jah242 4y ago
Contrarian hypothesis for you :) The more I work on AI/ML the more I think the fears around AGI have been misplaced. General intelligence almost by definition involves general motivations (rewards). Very general rewards inevitably make
25.
by jah242 4y ago
This would be one example from Deepmind using raw pixel input to stack objects. This has a relatively detailed reward function (but is also a very complicated task) - https://arxiv.org/abs/2110.06192 There are other ex
26.
by jah242 4y ago
There might be some truth in what you say for very large image and language models that use supervised learning. It is really hard to see how this 'it's just a lot of good data' view applies to deep reinforcement learning whe
27.
by jah242 4y ago
Sounds like we are in very similar positions and have a very similar question :). My only real plan so far is to try and beat or match SOTA on a recent benchmark from a large corporate / research lab, give them an email and hope they a
28.
by jah242 4y ago
Great video but I think what this really highlighted for me is how almost comically complicated ICE engines are to put together.
29.
by jah242 4y ago
Couldn't have asked for a better second instalment of the Fyre Festival franchise
30.
by jah242 4y ago
I would add that for the non-hospitalised groups it is substantially lower again at more like HR of 1.3-1.4 - (see supplementary tables - https://static-content.springer.com/esm/art%3A10.1038%2Fs415... ) I would also ad

More ›