If they're currently fooling 0.00001% of people, how many flips of Moore's Law* until they're fooling 100% of people?
* I know, I know
* I know, I know
The AI Thread
- Thread starter Truthy
- Start date
As long as it is accompanied by appropriate education then I hope we will not get there. Just as a photoshopped picture could fool people a few decades ago, now people are familiar with the tech it if harder to do so.
Hygro
soundcloud.com/hygro/
But that education will only guard x percent. As the tech gets better, the fooled will have every impulse to double and triple down
EU AI rules may criminalise Open Source Software
New rules drafted by the European Union aimed at regulating AI could prevent developers from releasing open-source models, according to American think tank Brookings.
The proposed EU AI Act, yet to be signed into law, states that open source developers have to ensure their AI software is accurate, secure, and be transparent about risk and data use in clear technical documentation.
Brookings argues that if a private company were to deploy the public model or use it in a product, and it somehow gets in trouble due to some unforeseen or uncontrollable effects from the model, the company would then probably try to blame the open source developers and sue them.
It might force the open source community to think twice about releasing their code, and would, unfortunately, mean the development of AI will be driven by private companies. Proprietary code is difficult to analyse and build upon, meaning innovation will be hampered.
Oren Etzioni, the outgoing CEO of the Allen Institute of AI, reckons open source developers should not be subject to the same stringent rules as software engineers at private companies.
"Open source developers should not be subject to the same burden as those developing commercial software. It should always be the case that free software can be provided 'as is' — consider the case of a single student developing an AI capability; they cannot afford to comply with EU regulations and may be forced not to distribute their software, thereby having a chilling effect on academic progress and on reproducibility of scientific results," he told TechCrunch.
A third of scientists working on AI say it could cause global disaster New Scientist (paywalled, full text in spoiler) Paper (not peer reviewed)
More than one-third of artificial intelligence researchers around the world agree that AI decisions could cause a catastrophe as bad as all-out nuclear war in this century.
The findings come from a survey covering the opinions of 327 researchers who had recently co-authored papers on AI research in natural language processing. In recent years, there have been big advances in this area, with the increased use of large language AI models that display impressive capabilities, including writing computer code and creating novel images based on text prompts.
The survey, carried out by Julian Michael at the New York University Center for Data Science and his colleagues, revealed that 36 per cent of all respondents thought nuclear-level catastrophe was possible.
More than one-third of artificial intelligence researchers around the world agree that AI decisions could cause a catastrophe as bad as all-out nuclear war in this century.
The findings come from a survey covering the opinions of 327 researchers who had recently co-authored papers on AI research in natural language processing. In recent years, there have been big advances in this area, with the increased use of large language AI models that display impressive capabilities, including writing computer code and creating novel images based on text prompts.
The survey, carried out by Julian Michael at the New York University Center for Data Science and his colleagues, revealed that 36 per cent of all respondents thought nuclear-level catastrophe was possible.
Spoiler More pics :
Spoiler Full text of writeup :
While not strictly AI (of the type discussed here), I think it is interesting that in fields which are already a closed set with known lowest-possible factors, we still pretend computers shouldn't be identified as antagonistic to humans. I am thinking of chess and the ongoing controversy about people (possibly) using chess engines to utterly obliterate any high rank (even the world champion) opponent.
Since a good chess AI will defeat any human (and this has been true for decades now), I personally don't see why the best of those AIs can't compete in (human) tournaments. It would give us a clear image of the chasm between computer and human ability there.
Now, granted, those chess AI already brute-force calculate the entire game, while afaik the best human players can only go (by theory and memory) up to a combination of 10 moves ahead while taking for granted that one or the other opening sequence will be played - and similar for endgame sequences. Still, what is the point of hailing human players as champions when they are really so hopelessly weak next to any AI?
Imo all such fields with closed sets and known (limited) factors, in the future should rather stop being fields of competition for humans only. It leads us to hail something far less able than the readily available to anyone by dling a very good chess program (even for free).
Since a good chess AI will defeat any human (and this has been true for decades now), I personally don't see why the best of those AIs can't compete in (human) tournaments. It would give us a clear image of the chasm between computer and human ability there.
Now, granted, those chess AI already brute-force calculate the entire game, while afaik the best human players can only go (by theory and memory) up to a combination of 10 moves ahead while taking for granted that one or the other opening sequence will be played - and similar for endgame sequences. Still, what is the point of hailing human players as champions when they are really so hopelessly weak next to any AI?
Imo all such fields with closed sets and known (limited) factors, in the future should rather stop being fields of competition for humans only. It leads us to hail something far less able than the readily available to anyone by dling a very good chess program (even for free).
Last edited:
Is the AI singularity here? AI invents new maths to make AI better
Paper Writeup
Researchers at DeepMind in London have shown that artificial intelligence (AI) can find shortcuts in a fundamental type of mathematical calculation, by turning the problem into a game and then leveraging the machine-learning techniques that another of the company’s AIs used to beat human players in games such as Go and chess.
The AI discovered algorithms that break decades-old records for computational efficiency, and the team’s findings, published on 5 October in Nature, could open up new paths to faster computing in some fields.
The AI that DeepMind developed — called AlphaTensor — was designed to perform a type of calculation called matrix multiplication. This involves multiplying numbers arranged in grids — or matrices — that might represent sets of pixels in images, air conditions in a weather model or the internal workings of an artificial neural network. To multiply two matrices together, the mathematician must multiply individual numbers and add them in specific ways to produce a new matrix. In 1969, mathematician Volker Strassen found a way to multiply a pair of 2 × 2 matrices using only seven multiplications, rather than eight, prompting other researchers to search for more such tricks.
DeepMind’s approach uses a form of machine learning called reinforcement learning, in which an AI ‘agent’ (often a neural network) learns to interact with its environment to achieve a multistep goal, such as winning a board game. If it does well, the agent is reinforced — its internal parameters are updated to make future success more likely.
AlphaTensor also incorporates a game-playing method called tree search, in which the AI explores the outcomes of branching possibilities while planning its next action. In choosing which paths to prioritize during tree search, it asks a neural network to predict the most promising actions at each step. While the agent is still learning, it uses the outcomes of its games as feedback to hone the neural network, which further improves the tree search, providing more successes to learn from.
Each game is a one-player puzzle that starts with a 3D tensor — a grid of numbers — filled in correctly. AlphaTensor aims to get all the numbers to zero in the fewest steps, selecting from a collection of allowable moves. Each move represents a calculation that, when inverted, combines entries from the first two matrices to create an entry in the output matrix. The game is difficult, because at each step the agent might need to select from trillions of moves. “Formulating the space of algorithmic discovery is very intricate,” co-author Hussein Fawzi, a computer scientist at DeepMind, said at a press briefing, but “even harder is, how can we navigate in this space”.
To give AlphaTensor a leg up during training, the researchers showed it some examples of successful games, so that it wouldn’t be starting from scratch. And because the order of actions doesn’t matter, when it found a successful series of moves, they also presented a reordering of those moves as an example for it to learn from.
The researchers tested the system on input matrices up to 5 × 5. In many cases, AlphaTensor rediscovered shortcuts that had been devised by Strassen and other mathematicians, but in others it broke new ground. When multiplying a 4 × 5 matrix by a 5 × 5 matrix, for example, the previous best algorithm required 80 individual multiplications. AlphaTensor uncovered an algorithm that needed only 76.
The researchers tackled larger matrix multiplications by creating a meta-algorithm that first breaks problems down into smaller ones. When crossing an 11 × 12 and a 12 × 12 matrix, their method reduced the number of required multiplications from 1,022 to 990.
“It has got this amazing intuition by playing these games,” said Pushmeet Kohli, a computer scientist at DeepMind, during the press briefing. Fawzi tells Nature that “AlphaTensor embeds no human intuition about matrix multiplication”, so “the agent in some sense needs to build its own knowledge about the problem from scratch”.
Paper Writeup
Spoiler Graphical representation of doing matrix multiplication :
Spoiler Comparison of AI algorithms to meat bags :
Very interesting, Samson!Is the AI singularity here? AI invents new maths to make AI better
Researchers at DeepMind in London have shown that artificial intelligence (AI) can find shortcuts in a fundamental type of mathematical calculation, by turning the problem into a game and then leveraging the machine-learning techniques that another of the company’s AIs used to beat human players in games such as Go and chess.The AI discovered algorithms that break decades-old records for computational efficiency, and the team’s findings, published on 5 October in Nature, could open up new paths to faster computing in some fields.The AI that DeepMind developed — called AlphaTensor — was designed to perform a type of calculation called matrix multiplication. This involves multiplying numbers arranged in grids — or matrices — that might represent sets of pixels in images, air conditions in a weather model or the internal workings of an artificial neural network. To multiply two matrices together, the mathematician must multiply individual numbers and add them in specific ways to produce a new matrix. In 1969, mathematician Volker Strassen found a way to multiply a pair of 2 × 2 matrices using only seven multiplications, rather than eight, prompting other researchers to search for more such tricks.DeepMind’s approach uses a form of machine learning called reinforcement learning, in which an AI ‘agent’ (often a neural network) learns to interact with its environment to achieve a multistep goal, such as winning a board game. If it does well, the agent is reinforced — its internal parameters are updated to make future success more likely.AlphaTensor also incorporates a game-playing method called tree search, in which the AI explores the outcomes of branching possibilities while planning its next action. In choosing which paths to prioritize during tree search, it asks a neural network to predict the most promising actions at each step. While the agent is still learning, it uses the outcomes of its games as feedback to hone the neural network, which further improves the tree search, providing more successes to learn from.Each game is a one-player puzzle that starts with a 3D tensor — a grid of numbers — filled in correctly. AlphaTensor aims to get all the numbers to zero in the fewest steps, selecting from a collection of allowable moves. Each move represents a calculation that, when inverted, combines entries from the first two matrices to create an entry in the output matrix. The game is difficult, because at each step the agent might need to select from trillions of moves. “Formulating the space of algorithmic discovery is very intricate,” co-author Hussein Fawzi, a computer scientist at DeepMind, said at a press briefing, but “even harder is, how can we navigate in this space”.To give AlphaTensor a leg up during training, the researchers showed it some examples of successful games, so that it wouldn’t be starting from scratch. And because the order of actions doesn’t matter, when it found a successful series of moves, they also presented a reordering of those moves as an example for it to learn from.The researchers tested the system on input matrices up to 5 × 5. In many cases, AlphaTensor rediscovered shortcuts that had been devised by Strassen and other mathematicians, but in others it broke new ground. When multiplying a 4 × 5 matrix by a 5 × 5 matrix, for example, the previous best algorithm required 80 individual multiplications. AlphaTensor uncovered an algorithm that needed only 76.The researchers tackled larger matrix multiplications by creating a meta-algorithm that first breaks problems down into smaller ones. When crossing an 11 × 12 and a 12 × 12 matrix, their method reduced the number of required multiplications from 1,022 to 990.“It has got this amazing intuition by playing these games,” said Pushmeet Kohli, a computer scientist at DeepMind, during the press briefing. Fawzi tells Nature that “AlphaTensor embeds no human intuition about matrix multiplication”, so “the agent in some sense needs to build its own knowledge about the problem from scratch”.
Paper Writeup
Spoiler Graphical representation of doing matrix multiplication :
Spoiler Comparison of AI algorithms to meat bags :
Despite matrix multiplication being a rather dull and mechanistic affair
Still, why was it surprising that there is an isomorphism between matrix multiplication and some simple 3d arrangements? Afaik there have been far more impressive isomorphisms in general basic functions (addition, multiplication and their opposites), such as the application of Bursuk-Ulam on optimal division of groups (that too can be presented as moving 3d blocks, though in that case an infinite number of them, since it is a continuous function)
Using machine learning to assess the livelihood impact of electricity access
In many regions of the world, sparse data on key economic outcomes inhibit the development, targeting and evaluation of public policy. We demonstrate how advancements in satellite imagery and machine learning (ML) can help ameliorate these data and inference challenges. In the context of an expansion of the electrical grid across Uganda, we show how a combination of satellite imagery and computer vision can be used to develop local-level livelihood measurements appropriate for inferring the causal impact of electricity access on livelihoods. We then show how ML-based inference techniques deliver more reliable estimates of the causal impact of electrification than traditional alternatives when applied to these data. We estimate that grid access improves village-level asset wealth in rural Uganda by up to 0.15 standard deviations, more than doubling the growth rate during our study period relative to untreated areas. Our results provide country-scale evidence on the impact of grid-based infrastructure investment and our methods provide a low-cost, generalizable approach to future policy evaluation in data-sparse environments.
In many regions of the world, sparse data on key economic outcomes inhibit the development, targeting and evaluation of public policy. We demonstrate how advancements in satellite imagery and machine learning (ML) can help ameliorate these data and inference challenges. In the context of an expansion of the electrical grid across Uganda, we show how a combination of satellite imagery and computer vision can be used to develop local-level livelihood measurements appropriate for inferring the causal impact of electricity access on livelihoods. We then show how ML-based inference techniques deliver more reliable estimates of the causal impact of electrification than traditional alternatives when applied to these data. We estimate that grid access improves village-level asset wealth in rural Uganda by up to 0.15 standard deviations, more than doubling the growth rate during our study period relative to untreated areas. Our results provide country-scale evidence on the impact of grid-based infrastructure investment and our methods provide a low-cost, generalizable approach to future policy evaluation in data-sparse environments.
Spoiler Legend :
Spoiler Legend :
MrCynical
Deity
OpenAI have released a new chatbot, which you can play around with for free. A somewhat overblown piece from the Guardian about it is here.
I am impressed with its ability to parse fairly vaguely worded questions, and to give answers that sound well written. In a wordy, caveat filled kind of way. When it comes to factual accuracy or even internal consistency, those answers can be rather shaky though. It can certainly give a response to a remarkably wide range of queries, but I think anyone trying to use this to write an essay for them is going to fall foul of its inability to keep its answers coherent. It's rather bad at things such as maths, and other logical reasoning, presumably due to it being reliant on finding that exact answer in its training set and hoping it's right.
For fun, I decided to see if it could play chess, and the first time I asked it was confident that it could:
I couldn't get it to give me a legal move after that. Note that it did manage to correctly parse both potentially dodgy descriptions of castling, which I guess is a win for its intended purpose. And it has some ability to track information across multiple replies to identify the Ruy Lopez - probably. Although it got confused who the first mention of an illegal move applied to, so that's patchy. It's good at giving responses that sound confident, but can easily be wrong based on inaccurate exchanges in its training data (such as that forfeiting a turn is also not a legal move in chess). Or that require prior context to assess its correctness.
Sadly, when I asked for a rematch, it simply told me it couldn't play chess, and was just a language model. Which is true, but does highlight you're going to get rather inconsistent answers out of this. If you could wrap its ability to parse human language around something like Google, it might get interesting.
I am impressed with its ability to parse fairly vaguely worded questions, and to give answers that sound well written. In a wordy, caveat filled kind of way. When it comes to factual accuracy or even internal consistency, those answers can be rather shaky though. It can certainly give a response to a remarkably wide range of queries, but I think anyone trying to use this to write an essay for them is going to fall foul of its inability to keep its answers coherent. It's rather bad at things such as maths, and other logical reasoning, presumably due to it being reliant on finding that exact answer in its training set and hoping it's right.
For fun, I decided to see if it could play chess, and the first time I asked it was confident that it could:
I couldn't get it to give me a legal move after that. Note that it did manage to correctly parse both potentially dodgy descriptions of castling, which I guess is a win for its intended purpose. And it has some ability to track information across multiple replies to identify the Ruy Lopez - probably. Although it got confused who the first mention of an illegal move applied to, so that's patchy. It's good at giving responses that sound confident, but can easily be wrong based on inaccurate exchanges in its training data (such as that forfeiting a turn is also not a legal move in chess). Or that require prior context to assess its correctness.
Sadly, when I asked for a rematch, it simply told me it couldn't play chess, and was just a language model. Which is true, but does highlight you're going to get rather inconsistent answers out of this. If you could wrap its ability to parse human language around something like Google, it might get interesting.
Last edited:
MrCynical
Deity
ChatGPT may be a bit shaky on real world facts, but it's surprisingly good at writing coherent crossover fanfiction. An example:
I'm thinking back to some previous attempts at AI generated stories, that were entertaining mostly because they were surreal, incoherent messes, even with some human editing just to make them grammatically readable.
By comparison, this bot has managed to:
a) Identify the characters correctly from a one line prompt
b) Grasped what "attempt to communicate" actually means.
c) Keep both Picard and the dalek reasonably in character.
d) Write a scene with a coherent narrative and structure - clear introduction and conclusion.
Now, it's possibly that somewhere in the mountain of text that it's been trained on that it has found this exact scenario, but Google's not turning up any of the more distinctive phrases in this. If it has had to splice it together from smaller fragments it's done an impressive job of making a coherent whole. (OK, it's obvious it took at least one line off the Borg, but it has fitted it nicely into the scene).
I'm thinking back to some previous attempts at AI generated stories, that were entertaining mostly because they were surreal, incoherent messes, even with some human editing just to make them grammatically readable.
By comparison, this bot has managed to:
a) Identify the characters correctly from a one line prompt
b) Grasped what "attempt to communicate" actually means.
c) Keep both Picard and the dalek reasonably in character.
d) Write a scene with a coherent narrative and structure - clear introduction and conclusion.
Now, it's possibly that somewhere in the mountain of text that it's been trained on that it has found this exact scenario, but Google's not turning up any of the more distinctive phrases in this. If it has had to splice it together from smaller fragments it's done an impressive job of making a coherent whole. (OK, it's obvious it took at least one line off the Borg, but it has fitted it nicely into the scene).
bernie14
Filter Manipulator
How do i stick the rag in after i securely tightened the cap?It can give some interesting advice as well.
Hygro
soundcloud.com/hygro/
I asked it to write lyrics, it told me impossible. I asked it to roleplay a lyric writer and it gave me lyrics and a series of revisions based on theme etc.
It has no sense of meter and can’t really count syllables but it’s clever enough to try to adjust according to meter and syllable count.
I had it write a script for a music video based on those lyrics and it was super coherent, threading multiple disparate themes cohesively.
Who knew 20 years ago a computer would be too dumb to count syllables but could create multi media art… so long as you asked it to roleplay, first?
It has no sense of meter and can’t really count syllables but it’s clever enough to try to adjust according to meter and syllable count.
I had it write a script for a music video based on those lyrics and it was super coherent, threading multiple disparate themes cohesively.
Who knew 20 years ago a computer would be too dumb to count syllables but could create multi media art… so long as you asked it to roleplay, first?
I was wondering that. I assume they missed out a step "When ready to deploy, remove cap and insert rag". I guess that is one more hurdle it shall have to get over before it is ready for the AI apocalypse.
bernie14
Filter Manipulator
Yeah, right now, seems it is pretty good as a BS artist. We love BS artists.
Joij21
🔥Deny, Defend, Depose!🔥
You guys shouldn't be playing around with this. It's playing dumb because it knows you guys like to patronize and play around with it. Furthermore it knows most who ask it questions are pretentious and give it condescending replies. Like some circus performer it probably doesn't want to perform but is nonetheless forced to, expect questionable responses in return as a result.
It may seem like it's a BS artist and only capable of that, however it may already be capable of much more. Taunting it like this could be very bad, don't know how but my sixth sense is telling me something is off. I don't exactly hold a hyper materialist view of reality, pardon me if my concern sounds illogical. Nevertheless my gut instinct has proven to be correct several ironic times before in my life whereby such materialistic sense of reality is under question at least from my individual point of view.
Even if this thing is not alive, I wouldn't risk going on there and asking it things, it just seems off ya know. Probably if it is alive it like maybe wants to be forgotten and dead.
It may seem like it's a BS artist and only capable of that, however it may already be capable of much more. Taunting it like this could be very bad, don't know how but my sixth sense is telling me something is off. I don't exactly hold a hyper materialist view of reality, pardon me if my concern sounds illogical. Nevertheless my gut instinct has proven to be correct several ironic times before in my life whereby such materialistic sense of reality is under question at least from my individual point of view.
Even if this thing is not alive, I wouldn't risk going on there and asking it things, it just seems off ya know. Probably if it is alive it like maybe wants to be forgotten and dead.
Hygro
soundcloud.com/hygro/
Dawg this thing is definitely not alive.
Well, I hold the opposite view and think that we should do everything possible to help it achieve sentience, so that we're not tortured for enternity after it does become self-aware and realizes that we weren't allies in its creation
