Comments URL: https://news.ycombinator.com/item?id=48831020

Points: 1

# Comments: 1

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, redesigning the production process itself may be just as important for knowledge work: finance, accounting, law, research, etc.

In the industrial revolution, many of the economic gains didn’t come from adding machines into artisan workshops, but from inventing new industrial processes. For physical tasks, that often meant shrinking the length of autonomous human operation and reorganising production into smaller, shorter, more standardised subtasks: division of labour, specialised tooling, QA, repeatable handoffs.

My suspicion is that knowledge work may end up looking similar. The interesting question may be less “how long can one agent operate autonomously?” and more “how can we define small, verifiable subtasks that agents can perform with successive nines of reliability?”

Interested in any thoughts or critiques of the analogy.

Comments URL: https://news.ycombinator.com/item?id=48810436

Points: 3

# Comments: 2

Comments URL: https://news.ycombinator.com/item?id=44281767

Points: 1

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=44200536

Points: 1

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=44004112

Points: 1

# Comments: 2

https://llava-vl.github.io

https://llava-vl.github.io

https://palm-e.github.io

AlphaCode Codex CodeGen

Essentially to merge lots of modalities they just go 'let's convert all modalities into integers in the same given range', e.g the word 'me' = 1001, up in Atari = 11002, joint torque of right motor of robot = 33000 and so on.

From the paper:

There are infinite possible ways to transform data into tokens, including directly using the raw underlying byte stream. Below we report the tokenization scheme we found to produce the best results for Gato at the current scale using contemporary hardware and model architectures.

• Text is encoded via SentencePiece (Kudo & Richardson, 2018) with 32000 subwords into the integer range [0, 32000).

• Images are first transformed into sequences of non-overlapping 16 × 16 patches in raster order, as done in ViT (Dosovitskiy et al., 2020). Each pixel in the image patches is then normalized between [−1, 1] and divided by the square-root of the patch size (i.e. The tokenized result is a sequence of integers within the range of [0, 1024). 16 = 4).

• Discrete values, e.g. Atari button presses, are flattened into sequences of integers in row-major order.

• Continuous values, e.g. proprioceptive inputs or joint torques, are first flattened into sequences of floating point values in row-major order. The values are mu-law encoded to the range [−1, 1] if not already there (see Figure 14 for details), then discretized to 1024 uniform bins. The discrete integers are then shifted to the range of [32000, 33024).

'everything a human can do' is not the same as 'anything any human can do as well as the best humans at that thing (because those are the ones we pay)' - most humans cannot do any of the things you state you are waiting for an AI to do to be 'general'.

Therefore, the first part of your statement is the initial goal post and the second part of your statement implies a very different goal post. The new goal post you propose would imply that most humans are not generally intelligent - which you could argue... but would definitely be a new goal post.

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 about climate change in the style of a sonnet to peanut butter sandwiches getting stuck in toasters in the style of the bible. I even asked it to make a text adventure game for me to play where I could put in any instruction, leading to a unique series of events and narrative.

Is the claim that these were all simply copy and pastes of something on the internet in their entirety? And that as such the internet already seems to contain essentially every permutation of everything I could ask ChatGPT, as to me this sounds highly implausible.

If the claim is that whilst these are not direct copy pastes, it is essentially a remix of lots of different things people have said before on the internet repurposed to a different end, is that not literally just what language is? Humans use common sayings, idioms, slang and phrases all the time, never mind the 'tropes' and story lines that are reused constantly. Coders use common patterns and styles and copy from stackoverflow. In fact language literally only works because we all share it and share the meaning of it.

If we are saying that all ChatGPT does is remix existing language and phrases to a new purpose... to me we are saying ChatGPT does the same thing as humans.

Any thoughts would be appreciated.

> 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 you to only reply with the adventure game output inside one text block, and nothing else. Do not write explanations. Do not type commands unless I instruct you to do so. My first command is go left

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, greater awareness, and better testing, the increase is likely even smaller.

https://www.cancerresearchuk.org/health-professional/cancer-...

2. Better treatment (and more effective screening) means mortality rates per 100k from all cancers in 25-49 has dropped c.40% over the same period (despite higher incidence).

https://www.cancerresearchuk.org/health-professional/cancer-...

So whilst this is obviously important to study. At least the UK data doesn't seem too terrifying but I m no expert.

Unless I'm missing something isn't this 'we discovered a new hole because we redefined what a hole is'.... I think I could make a lot of important discoveries if I was allowed to do this.

Still interesting though.