0:04
AI agents win at Slay the Spire 2 after researchers replace growing chat logs with structured memory
Researchers replaced the ever-growing chat log of AI agents with five separate memory layers, allowing them to process information more efficiently. This change kept the agent's prompt at around 5,000 tokens, a significant reduction from the ballooning 500,000 tokens it used to require. The structured memory approach enabled the AI to win 6 out of 10 games of Slay the Spire 2, a notable achievement given that competing agents couldn't even manage a single victory. The results suggest that organizing knowledge into distinct layers can be a key factor in improving AI performance, especially in complex decision-making tasks.
0:23
Grades dropped from 96 to 48 percent when a Brown professor made students take the exam without AI
He ran a take‑home exam and the class averaged 96 percent, then switched to a proctored, in‑person test. Overnight the average plummeted to just under 49 percent, and a handful of students dropped the course or vanished. The professor thinks most of the original scores were AI‑aided, and the sudden drop mirrors two large studies—one from China, one from UC Berkeley—that found students who rely on AI for homework see their exam scores collapse when the AI is removed. The pattern suggests the AI boost is real, but it also raises questions about how we assess learning when tools are so readily available.
0:42
OpenAI CEO Altman is now "pretty sure" AI is net job-creating, which is quite the pivot from predicting mass layoffs
Hey, I just read this thing that's got me thinking. It seems OpenAI's Sam Altman has done a complete 180 on AI's impact on jobs. He's now saying he's pretty sure it's created more jobs than it's eliminated. That's a huge shift from his earlier warnings about entire professions disappearing. I know he's not the only one who's been talking about this, but it's interesting to see him change his tune. Anthropic's CEO Dario Amodei is also walking back similar claims. The thing is, studies so far haven't really backed up either the doom-and-gloom predictions or the optimism. It's like, we're still figuring this out.
1:01
Claude Cowork's biggest use case is the mundane office work nobody wants to own, Anthropic says
Anthropic analyzed 1.2 million Claude Cowork sessions from more than 600,000 organizations. About half of all usage goes toward business processes and text creation, what Anthropic calls "the work around the work." That means tasks like compiling status reports, building onboarding checklists, or putting together slide decks. Software development barely shows up in Cowork because developers stick with Claude Code for that.
The article Claude Cowork's biggest use case is the mundane office work nobody wants to own, Anthropic says appeared first on The Decoder.
1:19
Meta kills Muse Image feature that let anyone generate AI photos of Instagram users without consent
Meta pulled a controversial feature from its new Muse Image model after widespread criticism. The feature let users generate AI images of other people by @-mentioning their public Instagram accounts. No consent needed, just a username. Meta admits "this feature missed the mark" and shut it down days after announcing it.
The article Meta kills Muse Image feature that let anyone generate AI photos of Instagram users without consent appeared first on The Decoder.
1:34
S&P Global sees OpenAI as a "key credit risk" for Oracle and cuts its credit rating
S&P Global has downgraded Oracle's credit rating to "BBB-," one notch above junk status. OpenAI accounts for roughly half of Oracle's $638 billion in contractual obligations. If OpenAI walked away, Oracle would be stuck with massive data center capacity it couldn't fill.
The article S&P Global sees OpenAI as a "key credit risk" for Oracle and cuts its credit rating appeared first on The Decoder.
