6 ms·

We still don't have squirrel-level AI. This is embarrassing. Now that LLMs have been around for a while, it's fairly clear what they can and can't do. There ar

We still don't have squirrel-level AI. This is embarrassing. Now that LLMs have been around for a while, it's fairly clear what they can and can't do. There are still some big pieces missing. Like some kind of world model.

p1esk 2y ago
fairly clear what they can and can't do It’s not at all clear what the next gen models will do (e.g. gpt5). Might be enough to trigger mass unemployment. Or not.
- DiscourseFan 2y ago
  Didn't OpenAI just cut its AGI department?
  - ben_w 2y ago
    AGI readiness team. With the department leader giving the statement that nobody is ready. Unless there's been another one since then? It's getting to be a bit of a blur.
    - DiscourseFan 2y ago
      Well, I'm also not ready to hear about my alzheimer's diagnoses, though I don't think that's coming any time soon.
  - pnut 2y ago
    That could mean the agi is taking over, for all we know.
    - DiscourseFan 2y ago
      Is AGI in the room with us right now?
- Animats 2y ago
  Bigger LLM models probably won't fix the underlying problems of hallucinations, lack of a confidence metric, and lack of a world model. They just do better at finding something relevant on already-solved problems.
  - p1esk 2y ago
    “Just do better” is the key metric here. 99% of all human jobs are “finding something relevant with already solved problems”.
JSteph22 2y ago
>There are still some big pieces missing. The most glaring one is that current LLMs are many, many orders of magnitude away from working on the equivalent of 900 calories per day of energy.
- Teever 2y ago
  I think you're correct that the energy efficiency of a human exceeds that of current computers, but I think it's a bit more complicated than a first order calorie count. How many joules go into producing those 900 calories? Like in terms of growing the food, from fertilizer production to tractor fuel, to feeding the farmer, to shipping the food, packaging it, storing it at the appropriate temperature, the ratio of spoiled food to actually consumed, the energy to cook it, all of that isn't counted in that simple 900 calorie measurement. I've been thinking about this for a while now but I haven't been able to quantify it so maybe someone reading this comment can help.
  - auggierose 2y ago
    I don't think it makes sense to include all of that in the calculation. A human doesn't need all of that, they just need 900 calories. You can just eat berries, no need to cook anything, for example.
    - amanaplanacanal 2y ago
      A human cannot live on just berries. Even if they could, where are those berries gonna come from?
      - auggierose 2y ago
        I am pretty sure they can. Berries grow in bushes, often.
        amanaplanacanal 2y ago
        Berries are missing essential amino acids, essential fatty acids, and essential vitamins. You could live on them for a while, I guess. Actually you can live without eating at all for a while. Both would lead to a painful death though.
nsbshssh 2y ago
There must be a name for the human bias that sees what is possible this month and doesn't believe things might be very different very soon and change quickly. AI has a tripple whamy that the models get more efficient, the chips will get faster and there will be more chips. Capitalist and government money pouring in. There is money to be made and it is national security. And to boot the cloud, big tech, cryptocurrency and gaming pouring more money into advancements in chips that boost ai.
- RaftPeople 2y ago
  > There must be a name for the human bias that sees what is possible this month and doesn't believe things might be very different very soon and change quickly The counter to that is that AI prediction of 10 to 20 years away has been happening since the 1950's (or before?). It seems there is some foundational progress but it's very slow.
  - nsbshssh 2y ago
    Unlike cold fusion there is real practical progress. No grid energy comes frok cold fusion. Millions of people are using AI to help get their work done.
    - RaftPeople 2y ago
      > Unlike cold fusion there is real practical progress. Yes agreed, there is progress, that's why I said "It seems there is some foundational progress but it's very slow." Training deep networks, vectorization of semantically related words+phrases, reservoir computing for time series patterns, etc. etc. All great foundational work, but we still can't reproduce how a rabbit dynamically adjusts it's olfactory network the minute it learns whether a new smell results in something positive/negative/neutral, which is probably the same trick used in a variety of places in our brain to give it such dynamic adaptability. It's progress, but it's moving slowly.
tim333 2y ago
AI development has been very uneven. Way better than squirrels at writing essays or playing chess. Way worse at climbing trees or finding nuts. I'm still waiting for an AI robot that can come to my house and fix an issue with the plumbing. Until that happens the terminator uprising is postponed.

p1esk 2y ago

fairly clear what they can and can't do It’s not at all clear what the next gen models will do (e.g. gpt5). Might be enough to trigger mass unemployment. Or not.

DiscourseFan 2y ago
Didn't OpenAI just cut its AGI department?
- ben_w 2y ago
  AGI readiness team. With the department leader giving the statement that nobody is ready. Unless there's been another one since then? It's getting to be a bit of a blur.
  - DiscourseFan 2y ago
    Well, I'm also not ready to hear about my alzheimer's diagnoses, though I don't think that's coming any time soon.
- pnut 2y ago
  That could mean the agi is taking over, for all we know.
  - DiscourseFan 2y ago
    Is AGI in the room with us right now?
Animats 2y ago
Bigger LLM models probably won't fix the underlying problems of hallucinations, lack of a confidence metric, and lack of a world model. They just do better at finding something relevant on already-solved problems.
- p1esk 2y ago
  “Just do better” is the key metric here. 99% of all human jobs are “finding something relevant with already solved problems”.

JSteph22 2y ago

>There are still some big pieces missing. The most glaring one is that current LLMs are many, many orders of magnitude away from working on the equivalent of 900 calories per day of energy.

Teever 2y ago
I think you're correct that the energy efficiency of a human exceeds that of current computers, but I think it's a bit more complicated than a first order calorie count. How many joules go into producing those 900 calories? Like in terms of growing the food, from fertilizer production to tractor fuel, to feeding the farmer, to shipping the food, packaging it, storing it at the appropriate temperature, the ratio of spoiled food to actually consumed, the energy to cook it, all of that isn't counted in that simple 900 calorie measurement. I've been thinking about this for a while now but I haven't been able to quantify it so maybe someone reading this comment can help.
- auggierose 2y ago
  I don't think it makes sense to include all of that in the calculation. A human doesn't need all of that, they just need 900 calories. You can just eat berries, no need to cook anything, for example.
  - amanaplanacanal 2y ago
    A human cannot live on just berries. Even if they could, where are those berries gonna come from?
    - auggierose 2y ago
      I am pretty sure they can. Berries grow in bushes, often.
      - amanaplanacanal 2y ago
        Berries are missing essential amino acids, essential fatty acids, and essential vitamins. You could live on them for a while, I guess. Actually you can live without eating at all for a while. Both would lead to a painful death though.

nsbshssh 2y ago

There must be a name for the human bias that sees what is possible this month and doesn't believe things might be very different very soon and change quickly. AI has a tripple whamy that the models get more efficient, the chips will get faster and there will be more chips. Capitalist and government money pouring in. There is money to be made and it is national security. And to boot the cloud, big tech, cryptocurrency and gaming pouring more money into advancements in chips that boost ai.

RaftPeople 2y ago
> There must be a name for the human bias that sees what is possible this month and doesn't believe things might be very different very soon and change quickly The counter to that is that AI prediction of 10 to 20 years away has been happening since the 1950's (or before?). It seems there is some foundational progress but it's very slow.
- nsbshssh 2y ago
  Unlike cold fusion there is real practical progress. No grid energy comes frok cold fusion. Millions of people are using AI to help get their work done.
  - RaftPeople 2y ago
    > Unlike cold fusion there is real practical progress. Yes agreed, there is progress, that's why I said "It seems there is some foundational progress but it's very slow." Training deep networks, vectorization of semantically related words+phrases, reservoir computing for time series patterns, etc. etc. All great foundational work, but we still can't reproduce how a rabbit dynamically adjusts it's olfactory network the minute it learns whether a new smell results in something positive/negative/neutral, which is probably the same trick used in a variety of places in our brain to give it such dynamic adaptability. It's progress, but it's moving slowly.

tim333 2y ago

AI development has been very uneven. Way better than squirrels at writing essays or playing chess. Way worse at climbing trees or finding nuts. I'm still waiting for an AI robot that can come to my house and fix an issue with the plumbing. Until that happens the terminator uprising is postponed.