What we're reading, in the order we're reading it.
Anthony and Harry's working stream — links, charts, tweets, and short takes. The unprocessed inputs behind the briefings.
(via DEV) Self-Adapting Language Models (via DEV)
Large language models (LLMs) are powerful but static; they lack mechanisms to adapt their weights in response to new tasks, knowledge, or examples. We introduce Self-Adapting LLMs (SEAL), a…
(via DEV) Google’s AI Is Actively Destroying the News Media (via DEV)
Google’s pivot to AI-powered search is proving disastrous for the digital news media landscape.
(via DEV) Design Patterns for Securing LLM Agents against Prompt Injections (via DEV)
This new paper by 11 authors from organizations including IBM, Invariant Labs, ETH Zurich, Google and Microsoft is an excellent addition to the literature on prompt injection and LLM security....
(via DEV) Google releases Weather Lab with AI-based cyclone predictions (via DEV)
Google is continuing its work in the area with a new Weather Lab website and experimental cyclone, or hurricane, predictions.
(via DEV) Disney and Universal lawsuit may be killing blow in AI copyright wars (via DEV)
Two huge movie studios are suing Midjourney, claiming the firm’s AI has been trained on their copyrighted material – the entrance of the Hollywood giants into this legal fight could...
(via DEV) Chatterbox TTS (via DEV)
Comments
(via DEV) Accidental Misalignment: Fine-Tuning LLMs Induces Unexpected Vulnerability (via DEV)
TL;DR This post discusses our explorations into the effects of domain-specific fine-tuning and how the characteristics of fine-tuning data relate to…
(via DEV) In landmark suit, Disney and Universal sue Midjourney for AI character theft (via DEV)
Multiple-studio complaint cites AI image outputs as evidence of “bottomless pit of plagiarism.”…
(via DEV) Have a damaged painting? Restore it in just hours with an AI-generated “mask” (via DEV)
A new method uses AI to physically restore a damaged painting much more quickly than what’s possible using manual techniques. A digitally generated “mask” in the form of thin film...
(via DEV) For Survivors Using Chatbots, ‘Delete’ Doesn’t Always Mean Deleted (via DEV)
Many survivors may assume that AI platforms and chatbots offer common privacy protections, but these are not guaranteed, Belle Torek writes.
(via DEV) AlphaOne gives AI developers a new dial to control LLM ‘thinking’ and boost performance (via DEV)
A new framework called AlphaOne is a novel way to modulate LLM thinking, improving model accuracy and efficiency without costly retraining.
(via DEV) Avoiding AI Deception: Lie Detectors can either Induce Honesty or Evasion (via DEV)
Large language models (LLMs) are often fine-tuned after training using methods like reinforcement learning from human feedback (RLHF). In this proces…
(via DEV) Meta AI’s experimental new smart glasses can see everything you do and even tell how you feel about it (via DEV)
Aria Gen 2 powers the brain inside the lenses
(via DEV) Manus has kick-started an AI agent boom in China (via DEV)
A stampede of contenders and copycats has followed the launch of the general AI agent, but China’s internet restrictions have so far made it easier to cater to overseas users.
(via DEV) How much information do LLMs really memorize? Now we know, thanks to Meta, Google, Nvidia and Cornell (via DEV)
Using a clever solution, researchers find GPT-style models have a fixed memorization capacity of approximately 3.6 bits per parameter.
(via DEV) Company boards push CEOs to replace IT workers with AI (via DEV)
The huge numbers of IT layoffs in 2024 and early 2025 are likely to continue as companies look to drive efficiencies with AI while bracing for a recession.
(via DEV) Stop guessing why your LLMs break: Anthropic’s new tool shows you exactly what goes wrong (via DEV)
Anthropic’s open-source circuit tracing tool can help developers debug, optimize, and control AI for reliable and trustable applications.
(via DEV) AGI Is Not Multimodal (via DEV)
“In projecting language back as the model for thought, we lose sight of the tacit embodied understanding that undergirds our intelligence.” –Terry Winograd The recent successes of generative AI models...
(via DEV) How vibe coding is tipping Silicon Valley’s scales of power (via DEV)
The rise of AI coding assistants puts software development into the hands of anyone with a good idea, with the next unicorn founder possibly having no technical background.
(via DEV) Rachel Thomas, PhD – Deep learning gets the glory, deep fact checking gets ignored (via DEV)
an AI researcher going back to school for immunology
(via DEV) Steering Vectors Can Help LLM Judges Detect Subtle Dishonesty (via DEV)
Code available at: https://github.com/watermeleon/judge_with_steered_response …
(via DEV) Phonely’s new AI agents hit 99% accuracy—and customers can’t tell they’re not human (via DEV)
Phonely, Maitai and Groq achieve breakthrough in AI phone support with sub-second response times and 99.2% accuracy, enabling human-level conversational AI for call centers.