Artificial Intelligence thread

[siegecrossbow]

Meanwhile..

https://twitter.com/i/web/status/1641892148084629504

Sounds like Twitter is a major security threat and should be banned under the Restrict Act. Their CEO often spouts Wumao propaganda on the platform too. The only question is whether 1 million dollar fine is sufficient punishment for one of the richest people in the world.

 

[Overbom]

Early days. Interesting to see what will come out in 6 or so months

https://twitter.com/i/web/status/1642181498278408193

 

[ougoah]

Please, Log in or Register to view URLs content!

I've no clue about this subject matter... at least the technical aspects and details. Just want to say this is the most interesting thread on SDF at the moment.

 

[Overbom]

Do we even know how general intelligence is quantified in humans, are we talking about an AI that never specialised in any field just continue adding data?

I think people want to prove that AI can't have intelligence for a b c reasons.

However, the LLMs models biggest revolution for the normal people is their commercial application. Do businesses really care if the AI they are using is AI with real intelligence instead of something close to it that almost-perfectly mimicks human "intelligence".

Maybe a more serious question, do most jobs that normal people have, really require advanced (analytical) intelligence that the AI can't mimick in a good-enough way?

I would wager that businesses are seeing GPT-4 and salivating, the CEOs, boardrooms, and shareholders are all getting ready for a big feast.

Is this going to happen today? No. Next year? Probably not. 2025, I expect yes, we will start seeing them used commercially in a viable and productive way. This field is absolutely exploding right now. Practically every week or so we have cutting edge research and real advancements.

Anyone else remembers the text-to-image tools from last year? They were useless, comically bad, ridiculous, actually mocked for being evidence against AI.. Check them out now, they are so good that, as we speak, the entire art community is staging a rebellion against them for "stealing" their art. Guess what, Adobe just created the same tool 2 weeks ago claiming that it was trained "legally" lol

Please, Log in or Register to view URLs content!

 

[BlackWindMnt]

Early days. Interesting to see what will come out in 6 or so months

https://twitter.com/i/web/status/1642181498278408193

Well if it can write unit/integration tests that can pass sonar-qube fuck sign me up i freaking hate writing that bullshit. But i don't expect that to be viable anytime soon maybe in the latter half of 2020s.

I think people want to prove that AI can't have intelligence for a b c reasons.

However, the LLMs models biggest revolution for the normal people is their commercial application. Do businesses really care if the AI they are using is AI with real intelligence instead of something close to it that almost-perfectly mimicks human "intelligence".

Maybe a more serious question, do most jobs that normal people have, really require advanced (analytical) intelligence that the AI can't mimick in a good-enough way?

I would wager that businesses are seeing GPT-4 and salivating, the CEOs, boardrooms, and shareholders are all getting ready for a big feast.

Is this going to happen today? No. Next year? Probably not. 2025, I expect yes, we will start seeing them used commercially in a viable and productive way. This field is absolutely exploding right now. Practically every week or so we have cutting edge research and real advancements.

Anyone else remembers the text-to-image tools from last year? They were useless, comically bad, ridiculous, actually mocked for being evidence against AI.. Check them out now, they are so good that, as we speak, the entire art community is staging a rebellion against them for "stealing" their art. Guess what, Adobe just created the same tool 2 weeks ago claiming that it was trained "legally" lol

Please, Log in or Register to view URLs content!

I think you're right in this case it just has to be good enough to convince the average customer, user etc.

 

[Overbom]

Getting one step closer

Please, Log in or Register to view URLs content!

In this work, we show that a pre-trained large language model (LLM) agent can execute computer tasks guided by natural language using a simple prompting scheme where the agent recursively criticizes and improves its output (RCI). The RCI approach significantly outperforms existing LLM methods for automating computer tasks and surpasses supervised learning (SL) and reinforcement learning (RL) approaches on the MiniWoB++ benchmark.

RCI is competitive with the state-of-the-art SL+RL method, using only a handful of demonstrations per task rather than tens of thousands, and without a task-specific reward function.

Furthermore, we demonstrate RCI prompting's effectiveness in enhancing LLMs' reasoning abilities on a suite of natural language reasoning tasks, outperforming chain of thought (CoT) prompting. We find that RCI combined with CoT performs better than either separately.

 

[xypher]

Getting one step closer

Please, Log in or Register to view URLs content!

Too computationally expensive on inference and models like GPT-4 are already bonkers in terms of compute demands, it is going to be easier to train a better model instead. The approach itself sounds similar to GANs in computer vision.

 

[xypher]

Meanwhile..

https://twitter.com/i/web/status/1641892148084629504

Chinese RecSys models are really good - e.g. look at TikTok's algorithm. That and computer vision are probably the strongest Deep Learning fields in China. For LLMs it seems that Chinese researchers are hitting the issue of data scarcity judging from the recent

Please, Log in or Register to view URLs content!

although the primary idea behind that paper was to showcase the HGC training of LLMs, so maybe it is not as bad. However, it seems like the next logical step for the Chinese NLP researchers would be to train multi-language model on both English & Chinese data to remedy the data issue.

 

[bobsagget]

I don't know much about software. But based on my observations of animals and humans, I have a very simple problem for judging whether AI is actually artificially intelligent or not: can it refuel itself?

Even the tiniest insect has this capability. Even the stupidest animal can use sensor fusion (via 3D imaging, trace concentration chemical sensors and acoustic detectors AKA eyes, nose and ears) to detect nutrients in 3D space, find the relative positions of itself and the nutrient, then vector towards that nutrient when reserves are low. More sophisticated predatory animals can use their sensor fusion to acquire an actively evading target and engage in autonomous combat to defeat the prey and acquire its nutrients.

Yet I don't see robots lining up at gas stations or plugging themselves in.

Wrong numerous automated devices self charge its such a simple task it indicates nothing.
Function when battery under 10 percent head to charging port or consume bio mass. Shit even spirit and opportunity had such functions with regard to solar collection. And a machine learning algo can do it to if allowed to learn. Gpt4 is so much more powerful and complicated

 

[FairAndUnbiased]

Wrong numerous automated devices self charge its such a simple task it indicates nothing.
Function when battery under 10 percent head to charging port or consume bio mass. Shit even spirit and opportunity had such functions with regard to solar collection. And a machine learning algo can do it to if allowed to learn. Gpt4 is so much more powerful and complicated

Can it do it against a resisting adversary in 3D space? Even a bug can do it.