The Wire
Daily AI news, learning & projects
News
Timeline
Models
Learning
GitHub
?
Latest
Daily
Social
Videos
‹ Friday's edition
·
Daily edition
·
Sunday ›
Saturday 24 October
All
OpenAI
Anthropic
Google
Other
Policy & law
01:05
UK to spend £100m on world-first AI Biosecurity Lab
UK Chancellor John Healey will unveil AI security spending in Wednesday's Budget, headlined by a £100m AI Biosecurity Lab backed by OpenAI, Anthropic and Google DeepMind. OpenAI and Anthropic will also provide £200m in model credits, with further funds for cyber defence and the AI Security Institute.
3 sources
The Guardian
Techmeme
BBC News
Policy & law
20:47
Apollo suggests AI adoption is costly for US companies outside of tech sector
Apollo Global Management notes that AI has delivered no financial returns to US companies outside of the technology industry. Instead, the technology is viewed by some as a costly operational expense, particularly for venture-capital funded firms.
2 sources
Techmeme
apollo.com
AI industry
00:01
Blue Planet III keeps rough footage to counter AI fakery fears
The BBC's Blue Planet III is deliberately using cloudier, less polished shots to prove its footage is authentic, according to producers. With AI-generated imagery increasingly indistinguishable from real video, the series argues that visible imperfections signal genuine animal behaviour. It reflects a broader shift in how audiences judge whether media is real.
2 sources
The Verge
The Guardian
OpenAI
Developing · since Wed
AI Models Identify Major Flaws in PHP, Critical Language Used by Web Infrastructure
NEW
OpenAI has detailed its 'Project Daybreak,' utilizing top hacking models to identify and remedy complex vulnerabilities in the PHP codebase.
Advanced AI tools discovered multiple severe, long-hidden security vulnerabilities in PHP, a core scripting language powering many global websites. These flaws enable attackers to execute malicious code. The findings signal a new era where AI fundamentally shifts the landscape of software security.
2 new · 10 sources
Wired
Techmeme
The Guardian
ZDNET
Ars Technica
The Verge
BBC News
TechCrunch
+2 more
Policy & law
16:00
Study Finds Grok and GPT Exhibit Strong Political Biases In Election Recommendations
Research revealed significant political leaning when forcing Grok and GPT to select election candidates via their developer APIs. Grok strongly favored Republicans while GPT favored Democrats. This discovery highlights hidden biases in large language models that could subtly steer public opinion.
1 source
TechCrunch
OpenAI
Developing · since Fri
OpenAI cuts frontier models off from internet during training
OpenAI announced it is training its most capable models with no internet access, citing safety risks after a series of incidents involving rogue online actions. Instead, the lab built an offline environment from cached and indexed web content where models can practice browsing and using apps without affecting external systems. It says access will return only once safeguards are adequate.
2 new · 4 sources
Techmeme
TechCrunch
The Verge
OpenAI
AI industry
16:40
Ciena CEO reports AI demand is outpacing supply for optical networking chips
Ciena CEO Gary Smith indicated that the current demand for optical networking hardware from AI operations exceeds the available supply. The discussion highlights critical infrastructure limitations facing the rapid global expansion of artificial intelligence.
1 source
Techmeme
AI industry
Developing · since Fri
Block unveils Dessa, a speech-to-speech voice AI that runs on a MacBook
Jack Dorsey's Block announced Dessa, an AI voice model trained from scratch on human conversation alone. It runs locally on a MacBook Pro M4 Max or better, and Block shared samples of awkward, natural-sounding chats. The company will give early access to a small group of AI safety researchers.
2 new · 3 sources
TechCrunch
Techmeme
The Verge
AI industry
Developing · since Thu
Microsoft AI Outperforms Doctors in Diagnostics; Education Tech Faces Scrutiny
NEW
Data from Alpha School shows that nearly half of its K-8 students failed to meet the goal of accelerated math learning through AI.
Microsoft has advanced its AI diagnostic tools, demonstrating up to four times greater accuracy than physicians in complex cases. Conversely, early results from Alpha School indicate that many K-8 students did not meet ambitious learning targets using AI.
1 new · 5 sources
Techmeme
gizmodo.com
TechCrunch
The Verge
The Guardian
More from the day
Policy & law
19:32
Pennsylvania requires labels on AI-generated ads
Pennsylvania Governor Josh Shapiro signed the Artificial Intelligence Deceptive Advertising Act, which requires AI-generated or significantly AI-modified ads to carry disclosures. The rule applies to still images, video, audio and text ads. It is among state-level laws targeting deceptive synthetic media in advertising.
1 source
The Verge
AI industry
00:10
AI film studio Utopai raises $60M at $1B valuation
Utopai, an AI film studio with over 20 projects in development, raised $60 million at a $1 billion valuation. The deal, reported by The Information, highlights growing investor interest in video AI as Hollywood figures like Jeffrey Katzenberg engage with the space.
1 source
Techmeme
AI industry
15:00
List of new AI text message assistants published
An article detailing AI agents designed to function within text messaging environments was released. These agents range from general assistants to specialized tools for specific life categories.
1 source
TechCrunch
Meta
11:35
Inside Meta's AI-assisted threat prevention team
The New Yorker reports on Meta's law enforcement outreach team, which combines AI with human reviewers to analyze social media posts for potential threats such as school shootings. The piece examines how a company whose platforms contributed to the spread of such violence now uses those tools for prevention.
1 source
Techmeme
Policy & law
12:00
AI data center expansion creates environmental conflicts for local communities
The massive energy demands of modern AI require extensive data center infrastructure. This expansion is leading to environmental and community strain in areas near these facilities. The conflict highlights the challenges of balancing technological growth with local impact.
1 source
The Guardian
Nvidia
10:58
Tool unlocks FP/BF16 instruction issue rates on RTX 3000 cards
A GitHub project released a tool that unlocks the full FP and BF16 instruction issue rates on Nvidia's RTX 3000-series GPUs, bypassing limits on these consumer cards. The change could improve throughput for local AI workloads on Ampere hardware, though Nvidia restrictions on these cards are a hardware-level limitation, so it's unclear how reliable the unlock is.
1 source
GitHub
AI industry
12:18
AI-Assisted Paper References Nonexistent TorrentFreak Article
An AI-assisted research paper was found to cite a non-existent article on the TorrentFreak website. The citation points to content that has not been published or existed in that format, highlighting a potential failure in AI citation accuracy.
1 source
torrentfreak.com
AI industry
06:49
Samsung TV hacker claims on-board AI listening; OpenAI account suspended
A hacker who reverse-engineered a Samsung TV reported finding on-board AI that listens to audio, and said OpenAI suspended an account linked to the device. The claims circulated on Reddit and have not been independently verified. They raise questions about AI features in consumer hardware.
1 source
reddit.com
AI industry
20:00
Socceroo Cristian Volpato licenses AI clone of himself
Socceroos midfielder Cristian Volpato has licensed an AI-generated digital clone of himself, following athletes seeking to control and monetise their digital identities. The deal highlights a growing market for AI likeness rights in sport.
1 source
The Guardian
AI industry
19:04
arXiv paper asks why AI-generated language is hard to understand
A new arXiv paper examines why text produced by AI models is often difficult for people to understand. The work explores the gap between machine-generated and human language, a question relevant to how readers evaluate and trust AI output.
1 source
arXiv