Microsoft introduces 365 Copilot: a new LLM based AI-copilot for the Microsoft Suite: Word, Excel, PowerPoint, Outlook, Teams. https://www.youtube.com/watch?v=Bf-dbS9CcRU
“RL agents🤖need a lot of data, which they usually need to gather themselves. But does that data need to be real? Enter *Synthetic Experience Replay*, leveraging recent advances in #GenerativeAI in order to vastly upsample⬆️ an agent’s training data!” https://twitter.com/cong_ml/status/1635642214586937346
"Mathew Lodge, the CEO of Diffblue, a company that uses reinforcement learning to automatically generate unit tests for Java code, said that “reinforcement systems alone are a fraction of the cost to run and can be vastly more accurate than LLMs, to the point that some can work with minimal human review.”…Sutskever said that as generative models improve, “they will have a shocking degree of understanding of the world and many of its subtleties, as seen through the lens of text.”" https://spectrum.ieee.org/ai-hallucination
Transformer-based World Models Are Happy With 100k Interactions https://arxiv.org/abs/2303.07109
Self-planning Code Generation with Large Language Model https://arxiv.org/abs/2303.06689
Matt Welsh’s vision of the future of programming: there isn’t one. Programming sucks, so let an AI do it. Humans write specifications (product managers), test and review automatically generated code, and train models to use new APIs. https://thenewstack.io/coding-sucks-anyway-matt-welsh-on-the-end-of-programming/
Map of AI existential safety https://aisafety.world/
“We're using a particle accelerator and AI to read a lost library from a dead empire. People have been trying to read the Herculaneum Papyri for 275 years. With your help, we'll do it in 2023.” https://twitter.com/natfriedman/status/1636028472819867650
China's answer to ChatGPT? Baidu shares tumble as Ernie Bot disappoints https://www.reuters.com/technology/chinese-search-giant-baidu-introduces-ernie-bot-2023-03-16/
No Evidence Against the Greater Male Variability Hypothesis: A Commentary on Harrison et al.’s (2022) Meta-Analysis of Animal Personality https://psyarxiv.com/6ua8r/
What does solar look like in the limit? https://twitter.com/TerraformIndies/status/1630481836676907009
“Hebrew legend imagined that in utopia, a single harvest was "sufficient for the needs of forty years" This is about exactly how cheap modern grains are Along material dimensions, we have created the heaven our ancestors dreamed of” https://twitter.com/ElonBachman/status/1630907694981824513
Can Policymakers Trust Forecasters? Experts, modelers, and forecasters try to predict events, but which of them are most reliable? https://progress.institute/can-policymakers-trust-forecasters/
Britain’s remarkable fight to end the slave trade https://www.spectator.co.uk/article/britains-remarkable-fight-to-end-the-slave-trade/
“Australia called for an inquiry into Covid origins. China immediately started coercing Australia with boycotts, trade and tourism restrictions.” https://twitter.com/whyvert/status/1636184375745290240
Russia's transfer of Ukrainian children constitutes a war crime, UN report says: "They are basically moving children from one group to another for the intentional erasure of their national or ethnic identity," https://www.semafor.com/article/03/16/2023/russia-transfer-ukraine-children-war-crime-un
GPT-4: Wow and Meh
Wow:
Meh:
Shit academics say: “This AI isn't really intelligent. It cannot reason. It does not understand what it is saying. It's just compressed data in the form of general algorithms being used to make predictions. Totally not Intelligence™.”
Why hasn't AI porn taken off yet? Possibly for the same reason that fully autonomous cars are still limited to a few cities. The technology is there, but even a small chance of a fatal error is too high. The porn industry might be scared shitless that their AI could accidentally generate images of children, or that someone might jailbreak it to do so.
When you started to compare porn and self-driving cars, I thought you were going to go in a different direction. I thought you were going to reference the last mile problem where getting from 98% accurancy to 99.9% accurancy takes more time than going from 0% to 98%.
To my eye, at least, the AI-generated photorealistic porn just isn't very good. And it's not just porn. All A.I-generated photorealistic depictions of people give me the heebie-jeebies. Everyone knows about the wrong numbers of fingers, but it's a lot of subtle things that are off: arms bending at unnatural angles, weird limb-to-torso proportions. It activates the part of my reptile brain that's supposed to depict diseased or disfigured humans.
On the other hand, the uncanny valley effect shouldn't effect cartoon/artistic pornography. And lo and behold, after perusing some AI-generated hentai, it's pretty good!