Crooked Bough News Round-up

— New study shows why simulated reasoning AI models don’t yet live up to their billing

 

at 4/25/25 3:43pm

Top AI models excel at math problems but lack reasoning needed for Math Olympiad proofs.

[Category: AI, Biz & IT, AI research, machine learning, simulated reasoning, SR models]

— FBI offers $10 million for information about Salt Typhoon members

 

at 4/25/25 2:38pm

FBI accepts tips by TOR in likely attempt to woo China-based informants.

[Category: Biz & IT, Policy, Security, china, hacking, salt typhoon, telecoms]

— In the age of AI, we must protect human creativity as a natural resource

 

at 4/25/25 5:00am

Op-ed: As AI outputs flood the Internet, diverse human perspectives are our most valuable resource.

[Category: AI, Biz & IT, Features, AI ethics, creative conservation, human creativity, machine learning]

— New Android spyware is targeting Russian military personnel on the front lines

 

at 4/24/25 2:02pm

Trojanized mapping app steals users' locations, contacts, and more.

[Category: Biz & IT, Policy, Security, android, malware, russia, spyware, Ukraine]

— Annoyed ChatGPT users complain about bot’s relentlessly positive tone

 

at 4/21/25 10:22am

Users complain of new "sycophancy" streak where ChatGPT thinks everything is brilliant.

[Category: AI, Biz & IT, AI behavior, AI sycophancy, ChatGPT, chatgtp, machine learning, openai, sycophancy]

— Company apologizes after AI support agent invents policy that causes user uproar

 

at 4/17/25 4:42pm

Frustrated software developer believed AI-generated message came from human support rep.

[Category: AI, Biz & IT, AI development tools, AI programming, chatbot, ChatGPT, chatgtp, cursor, Hacker News, machine learning, reddit]

— OpenAI releases new simulated reasoning models with full tool access

 

at 4/16/25 4:21pm

New o3 model appears "near-genius level," according to one doctor, but it still makes mistakes.

[Category: AI, Biz & IT, AI assistants, Anthropic, ChatGPT, chatgtp, Claude 3.7 Sonnet, gemini, Gemini 2.5 Pro, google, greg brockman, large language models, machine learning, openai, simulated reasoning, SR models]

— Researchers claim breakthrough in fight against AI’s frustrating security hole

 

at 4/16/25 5:15am

Prompt injections are the Achilles' heel of AI assistants. Google offers a potential fix.

[Category: AI, Biz & IT, Google, AI security, ChatGPT, chatgtp, google, google deepmind, large language models, machine learning, prompt injections, Riley Gooside, Simon Willison]

— 4chan has been down since Monday night after “pretty comprehensive own”

 

at 4/15/25 12:43pm

Early info is unreliable, but the site has been mostly unavailable for hours.

[Category: Biz & IT, Tech, 4chan]

— OpenAI continues naming chaos despite CEO acknowledging the habit

 

at 4/14/25 2:53pm

OpenAI's brand-new "GPT-4.1" has a funky name but reasonable performance for the price.

[Category: AI, Biz & IT, AI assistants, API, ChatGPT, chatgtp, GPT-4, GPT-4.1, GPT-4.5, large language models, machine learning, sam altman]

— Amid Trump tariff chaos, Nvidia launches AI chip production on US soil

 

at 4/14/25 9:55am

New announcement seems aimed at appeasing Trump, signals company shift.

[Category: AI, Biz & IT, Policy, AI chips, china, Chips Act, Donald Trump, GPUs, Jensen Huang, machine learning, NVIDIA, president trump, trump administration, US manufacturing]

— That groan you hear is users’ reaction to Recall going back into Windows

 

at 4/11/25 12:45pm

Snapshotting and AI processing a screen every 3 seconds. What could possibly go wrong?

[Category: AI, Biz & IT, Security, microsoft, privacy, recall, Windows]

— Researcher uncovers dozens of sketchy Chrome extensions with 4 million installs

 

at 4/11/25 5:15am

Even weirder: Why would Google give so many the "Featured" stamp for trustworthiness?

[Category: Biz & IT, Google, Security, chrome, Chrome Web Store, extensions, google]

— Researchers concerned to find AI models misrepresenting their “reasoning” processes

 

at 4/10/25 4:37pm

New Anthropic research shows AI models often fail to disclose reasoning shortcuts.

[Category: AI, Biz & IT, AI alignment, AI research, AI safety, Anthropic, ChatGPT, Claude, large language models, machine learning, simulated reasoning, SR models]

— OpenAI helps spammers plaster 80,000 sites with messages that bypassed filters

 

at 4/9/25 1:32pm

Company didn't notice its chatbot was being abused for (at least) 4 months.

[Category: Biz & IT, Security, AI, large language models, LLMs, openai, security, spam]

— After months of user complaints, Anthropic debuts new $200/month AI plan

 

at 4/9/25 1:20pm

Two-tiered "Claude Max" expands rate limits and offers traffic priority to subscribers.

[Category: AI, Biz & IT, AI assistants, Anthropic, ChatGPT, ChatGPT Pro, chatgtp, Claude, Claude Max, Claude Pro, large language models, machine learning, openai]

— “The girl should be calling men.” Leak exposes Black Basta’s influence tactics.

 

at 4/8/25 2:47pm

Disclosure of tactics, techniques, and procedures provides rare glimpse into secretive group.

[Category: Biz & IT, Security, exploits, ransomware, social engineering]

— Carmack defends AI tools after Quake fan calls Microsoft AI demo “disgusting”

 

at 4/8/25 12:26pm

Carmack: "Power tools" like AI are "central to all the progress in computers."

[Category: AI, Biz & IT, Gaming, AI criticism, AI development tools, game development, game synthesis, id Software, John Carmack, machine learning, microsoft, quake, quake ii, Tim Sweeney, X]

— Meta’s surprise Llama 4 drop exposes the gap between AI ambition and reality

 

at 4/7/25 1:54pm

Touted 10M token context proves elusive, while early performance tests disappoint experts.

[Category: AI, Biz & IT, LLaMA, Llama 3, Llama 4, machine learning, meta, Simon Willison]

— NSA warns “fast flux” threatens national security. What is fast flux anyway?

 

at 4/4/25 2:17pm

Used by nation-states and crime groups, fast flux bypasses many common defenses.

[Category: Biz & IT, Security, fast flux]

— Gmail unveils end-to-end encrypted messages. Only thing is: It’s not true E2EE.

 

at 4/3/25 3:16pm

Yes, encryption/decryption occurs on end-user devices, but there's a catch.

[Category: Biz & IT, Google, Security, encryption, gmail, google]

As of 4/25/25 9:10pm. Last new 4/25/25 5:54pm.