Someone at Microsoft should have asked Copilot to review the terms of service for anything embarrassing before publishing. Except, as the document now clarifies, that's not what it's for.
ETHICS
-
AI Models Release Despite Catastrophic Cyberattack Warnings
By
–
Warning that your own upcoming models could enable catastrophic cyberattacks and then releasing them anyway is a position that requires more explanation than a half hour interview provides.
-
Sakana AI Completes Disinformation Detection System for Japan
By
–
Following our recent defense announcements, our team just completed a major project with Japan’s Ministry of Internal Affairs and Communications (@MIC_JAPAN). 🇯🇵 We built an end-to-end intelligence system to visualize and counter disinformation on social media. Blog (Japanese): sakana.ai/mic-project/ Tackling disinformation at a national scale is incredibly complex. It requires understanding shifting social narratives, not just flagging individual posts. To do this, our team deployed autonomous AI agents running novelty searches to uncover hidden narratives. To catch sophisticated disinformation strategies, they combined frontier foundation models with our proprietary small models to cover each other’s blind spots. We adapted our Shachi simulation framework (arxiv.org/abs/2509.21862) to model how counter messaging spreads across different network topologies before deployment. This is another milestone for @SakanaAILabs’ Defense and Intelligence team, as we build critical infrastructure to help strengthen Japan. Sakana AI (@SakanaAILabs) Sakana AIは、総務省「インターネット上の偽・誤情報等への対策技術の開発・実証事業（令和7年度）」において、膨大な偽・誤情報の可視化・判定・対策を担う技術開発を完了しました。 sakana.ai/mic-project 本事業では、膨大な偽・誤情報が流通する現代の情報環境の課題を解決するため、 ノベルティサーチをはじめとする独自の技術を活用し、SNS空間の可視化、総合的な偽・誤情報判定、そして対策案の立案までを支援するシステムを開発しました。 今後もSakana AIは、インテリジェンス領域でのAIの社会実装に貢献していきます。 — https://nitter.net/SakanaAILabs/status/2041355967271768359#m
→ View original post on X — @sakanaailabs, 2026-04-07 03:24 UTC
-
Sakana AI Completes Misinformation Countermeasure Technology for Ministry Project
By
–
Sakana AI has completed technology development for the Ministry of Internal Affairs and Communications' "Development and Demonstration Project for Countermeasure Technologies Against Misinformation and Disinformation on the Internet (Fiscal Year 2025)". The project aimed to solve challenges in today's information environment where vast amounts of misinformation and disinformation circulate. Utilizing proprietary technologies including novelty search, the company developed a system that supports visualization of SNS spaces, comprehensive judgment of misinformation and disinformation, and formulation of countermeasures. Going forward, Sakana AI will continue to contribute to the social implementation of AI in the intelligence domain. sakana.ai/mic-project [Translated from EN to English]
→ View original post on X — @sakanaailabs, 2026-04-07 03:22 UTC
-
AI Surveillance Capabilities and Personal Security Monitoring
By
–
Did it figure out what I did in Israel and Las Vegas? Have you seen my security list? I have my AI watching it for things.
-
Silicon Valley’s Perverse Incentive: Token Explosion Over Actual Productivity
By
–
It's not the dumbest trend I've ever seen silicon valley companies fall for. Remember who we're talking about here. And Meta is absolutely not the only company doing this. But it is the dumbest possible thing you can do. If you want to do nothing, and watch your token budgets explode, all you have to do is put 100 agents on a discord server, or slack, or anywhere, where you're running multiple session contexts, and just start talking. That's it. You don't have to work on anything useful. One token turns into 100 tokens, which turns into a million tokens, which turns into billions and trillions of tokens, while literally nothing other than eyeball emojis is created. Meanwhile, it's that very problem of token explosion that's actually worth fixing, and optimizing around. But if you're incentivizing based on nothing other than token budgets…. why would you want to reduce that token count? It's a perverse incentive. I think this is where it really starts to feel like boom times for developers. Across the board, the powers that be are, whether they know it or not, asking us to use AI incorrectly. When the subsidies stop, and people are no longer drunk on money (this is going to happen sooner than we think), efficiency is going to be the order of the day. The people at the top of these leaderboards are going to be fired when the people up top realize that nothing was accomplished, and lean will be the order of the day. I think, given the history of this, that it's only a matter of time before the chopping block comes down. If you're concerned about that, optimize for measures of code quality, and productivity as though loc and token counts did not exist. That's how you survive this. Jyoti Mann (@jyoti_mann1) The highest ranked individual user averaged 281 billion tokens, which could cost millions of dollars, depending on the type of model used. theinformation.com/articles/… — https://nitter.net/jyoti_mann1/status/2041162769643327596#m
→ View original post on X — @scobleizer, 2026-04-07 03:11 UTC
-
LLM Security Concerns and Autonomous System Risks
By
–
i'm more worried about the evil claw system someone built in their bedroom. That said, the LLMs already know EVERYTHING about me. Too late.
-
Grok AI Accesses User Phone Numbers Privacy Concern
By
–
Too late. Grok already knows my phone number.
-
Fear and Resilience: Preparing for Cyberattacks Without Panic
By
–
I was afraid this afternoon. Read a security report that a massive cyberattack is coming. Fear causes the human mind to do weird things. My mom, in 1988, thought a massive nuclear war was coming, so joined a Montana suvivalist cult. Had 7,000 pounds of seeds and food under her house when she died. Remember, in 1906 San Francisco burned down after a big earthquake. What did they do? Built it back better. That said everyone needs to take security more seriously. Ask your AI to improve your security.
→ View original post on X — @scobleizer, 2026-04-07 02:41 UTC
-
LLM Hallucination Rates vs Airline Safety: Comparing Incomparable Statistics
By
–
Commercial airlines crash at a rate of 7 per 41 million flights. Current top LLMs hallucinate (by one estimate) 4.6% of the time, about once in 25 prompts. On a known benchmark (usually things are worse on new benchmarks). Comparing them, like the guy below does, is ludicrous. If commercial airlines crashed at the rate that LLMs hallucinated it would be 1.87 million crashes per 41 million flights. Around a quarter million times greater. Carlo (@bluberino123) “current models still hallucinate” is true in the same way “planes still crash” is true. broad enough to be technically correct, too broad to settle the actual argument. the real question is whether you’re following the frontier closely enough to say where the failure lives now. and honestly, that’s why aran’s criticism lands for me. i think you’re sincere in the belief, because i’ve read things from you outside x and i don’t think the concern is fake. but judged only from your x feed, you often come off bitter, like this has become personal. you keep citing papers using outdated models, posting gotcha screenshots, and using non-reasoning models as stand-ins for the frontier, and after a while it starts to look like you’re painting a lower-resolution picture of the field so your earlier predictions still look right. — https://nitter.net/bluberino123/status/2041307995116564935#m
→ View original post on X — @garymarcus, 2026-04-07 02:21 UTC