Technological advances raise new puzzles and challenges for cognitive science and the study of how humans think about and interact with artificial intelligence (AI). For example, the advent of large language models and their human-like linguistic abilities has raised substantial debate regarding whether or not AI could be conscious. Here, we consider the question of whether AI could have subjective experiences such as feelings and sensations (‘phenomenal consciousness’). While experts from many fields have weighed in on this issue in academic and public discourse, it remains unknown whether and how the general population attributes phenomenal consciousness to AI. We surveyed a sample of US residents (n = 300) and found that a majority of participants were willing to attribute some possibility of phenomenal consciousness to large language models. These attributions were robust, as they predicted attributions of mental states typically associated with phenomenality—but also flexible, as they were sensitive to individual differences such as usage frequency. Overall, these results show how folk intuitions about AI consciousness can diverge from expert intuitions—with potential implications for the legal and ethical status of AI. — Read More
Recent Updates Page 104
AI supercharges data center energy use – straining the grid and slowing sustainability efforts
The artificial intelligence boom has had such a profound effect on big tech companies that their energy consumption, and with it their carbon emissions, have surged.
The spectacular success of large language models such as ChatGPT has helped fuel this growth in energy demand. At 2.9 watt-hours per ChatGPT request, AI queries require about 10 times the electricity of traditional Google queries, according to the Electric Power Research Institute, a nonprofit research firm. Emerging AI capabilities such as audio and video generation are likely to add to this energy demand
The energy needs of AI are shifting the calculus of energy companies. They’re now exploring previously untenable options, such as restarting a nuclear reactor at the Three Mile Island power plant that has been dormant since the infamous disaster in 1979.
Data centers have had continuous growth for decades, but the magnitude of growth in the still-young era of large language models has been exceptional. AI requires a lot more computational and data storage resources than the pre-AI rate of data center growth could provide. — Read More
Generating audio for video
Video-to-audio research uses video pixels and text prompts to generate rich soundtracks
Video generation models are advancing at an incredible pace, but many current systems can only generate silent output. One of the next major steps toward bringing generated movies to life is creating soundtracks for these silent videos.
Today, we’re sharing progress on our video-to-audio (V2A) technology, which makes synchronized audiovisual generation possible. V2A combines video pixels with natural language text prompts to generate rich soundscapes for the on-screen action. — Read More
Read the Paper
Better & Faster Large Language Models via Multi-token Prediction
Large language models such as GPT and Llama are trained with a next-token prediction loss. In this work, we suggest that training language models to predict multiple future tokens at once results in higher sample efficiency. More specifically, at each position in the training corpus, we ask the model to predict the following n tokens using n independent output heads, operating on top of a shared model trunk. Considering multi-token prediction as an auxiliary training task, we measure improved downstream capabilities with no overhead in training time for both code and natural language models. The method is increasingly useful for larger model sizes, and keeps its appeal when training for multiple epochs. Gains are especially pronounced on generative benchmarks like coding, where our models consistently outperform strong baselines by several percentage points. Our 13B parameter models solves 12 % more problems on HumanEval and 17 % more on MBPP than comparable next-token models. Experiments on small algorithmic tasks demonstrate that multi-token prediction is favorable for the development of induction heads and algorithmic reasoning capabilities. As an additional benefit, models trained with 4-token prediction are up to 3 times faster at inference, even with large batch sizes. — Read More
What is AI?
Everyone thinks they know but no one can agree. And that’s a problem
AI is sexy, AI is cool. AI is entrenching inequality, upending the job market, and wrecking education. AI is a theme-park ride, AI is a magic trick. AI is our final invention, AI is a moral obligation. AI is the buzzword of the decade, AI is marketing jargon from 1955. AI is humanlike, AI is alien. AI is super-smart and as dumb as dirt. The AI boom will boost the economy, the AI bubble is about to burst. AI will increase abundance and empower humanity to maximally flourish in the universe. AI will kill us all.
What the hell is everybody talking about? — Read More
Perforation-type anchors inspired by skin ligament for robotic face covered with living skin
Skin equivalent, a living skin model composed of cells and extracellular matrix, possesses the potential to be an ideal covering material for robots due to its biological functionalities. To employ skin equivalents as covering materials for robots, a secure method for attaching them to the underlying structure is required. In this study, we develop and characterize perforation-type anchors inspired by the structure of skin ligaments as a technique to effectively adhere skin equivalents to robotic surfaces. To showcase the versatility of perforation-type anchors in three-dimensional (3D) coverage applications, we cover a 3D facial mold with intricate surface structure with skin equivalent using perforation-type anchors. Furthermore, we construct a robotic face covered with dermis equivalent, capable of expressing smiles, with actuation through perforation-type anchors. With the above results, this research introduces an approach to adhere and actuate skin equivalents with perforation-type anchors, potentially contributing to advancements in biohybrid robotics. — Read More
Why AI can’t replace science
The scientific revolution has increased our understanding of the world immensely and improved our lives immeasurably. Now, many argue that science as we know it could be rendered passé by artificial intelligence.
… Today, AI is being increasingly integrated into scientific discovery to accelerate research, helping scientists generate hypotheses, design experiments, gather and interpret large datasets, and write papers. But the reality is that science and AI have little in common and AI is unlikely to make science obsolete. The core of science is theoretical models that anyone can use to make reliable descriptions and predictions. … The core of AI, in contrast, is data mining. … However, without an underlying causal explanation, we don’t know whether a discovered pattern is a meaningful reflection of an underlying causal relationship or meaningless serendipity. — Read More
Perplexity’s grand theft AI
In every hype cycle, certain patterns of deceit emerge. In the last crypto boom, it was “ponzinomics” and “rug pulls.” In self-driving cars, it was “just five years away!” In AI, it’s seeing just how much unethical shit you can get away with.
Perplexity, which is in ongoing talks to raise hundreds of millions of dollars, is trying to create a Google Search competitor. Perplexity isn’t trying to create a “search engine,” though — it wants to create an “answer engine.” The idea is that instead of combing through a bunch of results to answer your own question with a primary source, you’ll simply get an answer Perplexity has found for you. “Factfulness and accuracy is what we care about,” Perplexity CEO Aravind Srinivas told The Verge.
That means that Perplexity is basically a rent-seeking middleman on high-quality sources. The value proposition on search, originally, was that by scraping the work done by journalists and others, Google’s results sent traffic to those sources. But by providing an answer, rather than pointing people to click through to a primary source, these so-called “answer engines” starve the primary source of ad revenue — keeping that revenue for themselves. Perplexity is among a group of vampires that include Arc Search and Google itself. — Read More
Polynomial Time Cryptanalytic Extraction of Neural Network Models
Billions of dollars and countless GPU hours are currently spent on training Deep Neural Networks (DNNs) for a variety of tasks. Thus, it is essential to determine the difficulty of extracting all the parameters of such neural networks when given access to their black-box implementations. Many versions of this problem have been studied over the last 30 years, and the best current attack on ReLU-based deep neural networks was presented at Crypto’20 by Carlini, Jagielski, and Mironov. It resembles a differential chosen plaintext attack on a cryptosystem, which has a secret key embedded in its black-box implementation and requires a polynomial number of queries but an exponential amount of time (as a function of the number of neurons)
In this paper, we improve this attack by developing several new techniques that enable us to extract with arbitrarily high precision all the real-valued parameters of a ReLU-based DNN using a polynomial number of queries and a polynomial amount of time. We demonstrate its practical efficiency by applying it to a full-sized neural network for classifying the CIFAR10 dataset, which has 3072 inputs, 8 hidden layers with 256 neurons each, and about 1.2 million neuronal parameters. An attack following the approach by Carlini et al. requires an exhaustive search over 2256 possibilities. Our attack replaces this with our new techniques, which require only 30 minutes on a 256-core computer. — Read More
Rick's Cafe AI 