D.A.D. joke of the day: People worry AI will replace workers, but mine fits right in with the team: brilliant, tireless, and does absolutely nothing until it's prompted.
What's New
AI developments from the last 24 hours
AI Model Prices Keep Falling as OpenAI, Anthropic Compete
OpenAI and Anthropic have been locked in a price war, developer Max Woolf writes on his blog, cutting costs on their AI models by as much as 80% over a few months while continuing to release upgraded versions. The steepest cuts target 'cache reads'—the discounted rate for reprocessing repeated context, which makes up the bulk of spending in coding-assistant workloads. Anthropic's Opus 5.5 now costs 40-60% less than its predecessor; OpenAI's newest models launched at roughly half the price of the versions they replaced.
Why it matters: If you or your team pay for AI coding tools or build on these models, your per-task costs are falling sharply even as capability holds steady—a rare case where competition is delivering both cheaper and better at the same time.
Discuss on Hacker News · Source: minimaxir.com
IBM CEO Says Businesses Can Skip ChatGPT Fees With In-House AI
One reason the big labs are cutting prices (see above): the free alternatives keep getting better. IBM's CEO says open weight AI models—freely downloadable systems like Llama or China's Qwen and DeepSeek—are now good and cheap enough that businesses can run them on their own servers instead of paying for ChatGPT or Claude subscriptions. He suggested most companies could get by with just two or three server racks. The catch, by his own admission: most enterprises lack the in-house expertise to wire up the GPUs, networking, and caching needed to actually run them. Commenters noted IBM sells that hardware, so the pitch isn't disinterested.
Why it matters: If true even partly, it changes the calculus for any company weighing AI subscription costs against building owned infrastructure—but the skills gap IBM's own CEO flagged may matter more than the hardware price.
Discuss on Hacker News · Source: theregister.com
Old Specialized AI Still Beats Chatbots at Diagnosing Skin Cancer
A dermatologist pitted a decade-old, purpose-built image classifier against GPT-6 Astra and Claude Fable 5.1 on 200 confirmed skin biopsies, asking each to pick the correct diagnosis from six options. The old model, fine-tuned for roughly twenty minutes, won decisively: 62% accuracy versus 42% for both chatbots, and it caught more melanomas. The LLMs also made costly errors in opposite directions—GPT-6 Astra over-diagnosed basal cell carcinoma, misreading 14 of 30 melanomas as that instead, while Claude Fable 5.1 over-called actinic keratosis.
Why it matters: General-purpose chatbots are being used informally for medical image questions, but this suggests narrow, specialized AI still beats frontier models at the specific visual task it was built for—a caution against assuming bigger and newer always means better.
Discuss on Hacker News · Source: lynchlab.net
What's Controversial
Stories sparking genuine backlash, policy fights, or heated disagreement in the AI community
Creative Industries Lost 200,000 Jobs in Four Years, AI Likely a Factor
US arts and media employment has fallen by more than 200,000 jobs since mid-2022, to about 1.2 million, with roughly 50,000 lost in the past year, economics writer Joseph Politano calculates from Bureau of Labor Statistics data on his Apricitas blog. For scale, the whole US economy added only about 45,000 jobs a month over the past year. Nearly half the losses are in film, TV and sound recording, where employment is down more than 30%, near a 30-year low. The telling pattern: since ChatGPT launched, every digital creative field, from graphic design to publishing, has shrunk, while in-person arts like museums and live performance keep growing. Politano says AI's exact share is hard to prove, partly because studios are cagey about using it, and because Hollywood consolidation, production moving overseas and the rise of social-media creators are hitting at the same time.
Why it matters: Arts jobs are a small slice of the economy, but Politano sees them as a preview for larger "creative-adjacent" white-collar work. Tech, where coding agents are spreading fastest, is in its longest hiring slump on record. His worst case for media is what the internet did to newspapers, whose payrolls are down nearly 85%. But he cautions that technology's effects are hard to predict: book sales have jumped 70% in six years, thanks to TikTok. Richard Florida, who coined the term "creative class," has questioned the numbers.
Discuss on Hacker News · Sources: Apricitas (Joseph Politano) · Fortune via inkl (Richard Florida) · BLS
Elvis's Music Owners Sue Anthropic Over 274 Songs
The owners of Elvis Presley's music, led by Authentic Brands Group, sued Anthropic on Friday, accusing it of using 274 of his songs, including "Can't Help Falling in Love" and "Don't Be Cruel," to train its Claude models without permission, The Hollywood Reporter reports, citing Billboard. They are seeking up to $150,000 per song, about $41 million. "The Elvis Plaintiffs bring this action to enforce these exclusive rights against the defendant that has built its entire business model based on copyright infringement at a scale never seen before," wrote lead attorney Richard Busch. Anthropic told Pitchfork it disagrees with the claims and will fight the case.
Why it matters: It adds to Anthropic's copyright fights. The company paid a record $1.5 billion to settle with book authors (D.A.D., July 22), and music publishers including Universal Music Group are separately suing it for $3 billion over some 20,000 works. As Anthropic prepares to go public, these cases are a measure of what AI companies may still owe for the material they trained on.
Sources: The Hollywood Reporter · AV Club · TechCrunch (music publishers' suit)
What's in the Lab
New announcements from major AI labs
Anthropic Offers Free Claude Access to Life Sciences Researchers
Anthropic is opening applications for guest researchers in life sciences, offering academic, nonprofit, and government labs free access to Claude plus 8-12 weeks working alongside Anthropic staff. Target areas include human biology, molecular biology, medicinal chemistry, drug discovery, and clinical science. Selected investigators keep rights to their own data and follow-on work, and results are meant to be published and open-sourced where possible. It follows Anthropic's recent moves into health and drug discovery, including its scorpion-toxin research (D.A.D., October 21).
Why it matters: It's part of a broader race among AI labs to prove their models belong in serious lab science, and to win over academic researchers while doing it.
What's in Academe
New papers on AI and its effects from researchers
AI Models Still Trail Traditional Tools at Predicting Household Spending
A new benchmark called HouseholdBench tests whether AI models can predict real household economic behavior—spending, income, job decisions, and how families might respond to policy changes—using data from six major U.S. surveys. Researchers found most large language models beat a simple no-change prediction, but traditional statistical models (gradient-boosted trees) still generally outperformed them. Notably, a small open-weight model fine-tuned on the data and set to generate multiple predictions per question matched larger proprietary models, even on policy scenarios it wasn't trained on.
Why it matters: Economists and policymakers often rely on surveys and statistical models to forecast how households will react to tax changes, rate hikes, or subsidies—this research suggests AI could become a cheaper, faster supplementary tool, though it's not yet ready to replace established forecasting methods.
AI Analysis of Earnings Calls Could Predict Economic Trends Early
Researchers fed generative AI more than 120,000 corporate earnings call transcripts to gauge what executives really expect for their businesses, building what they call an 'AI Economy Score.' The score reportedly forecasts GDP growth, production, employment and wages five or more quarters out—outperforming traditional economist surveys and existing forecasting tools. The effect is strongest when drawing on managers with track records of accurate, less-biased predictions, suggesting executives' own words carry signal that standard surveys miss.
Why it matters: If the findings hold up, it suggests AI can mine routine corporate disclosures for economic forecasts that rival or beat professional economists—a potential new tool for investors, policymakers and business planners trying to see around the next corner.
Study Finds Pay, Not Oversight, Keeps Workers Training Their AI Replacements
A new economic model from researchers Joshua Gans and Peter Landry examines what happens when firms ask workers to train the AI systems meant to replace them. Since firing a worker who sabotages training no longer carries the same threat once the job is ending anyway, the usual tool bosses use to prevent slacking stops working. The fix, the model shows, isn't more oversight—it's money: paying workers enough during training that cooperating pays better than quietly dragging it out, plus a bonus for finishing the job.
Why it matters: As more companies ask employees to help automate their own roles, this suggests the economically rational move is a severance-style completion bonus, not surveillance—a framework worth watching as real-world AI transition packages start taking shape.
Few Consumers Pay for AI Chatbots, and Most Switch Often
A new NBER study mined credit and debit card data from 174 million consumers to measure who actually pays for AI chatbots. Just 2.9% of U.S. consumers pay for a service like ChatGPT, Claude, or Gemini, almost all on cheap base-tier plans under $30/month. Annual customer retention runs only 20% to 45%—worse than most subscription products—and switching spikes whenever a rival releases a new model. The researchers conclude consumers see little difference between providers and will keep chasing the best deal rather than staying loyal to one brand.
Why it matters: If users treat AI chatbots as interchangeable commodities rather than loyal relationships, it undercuts the pricing power and lock-in that justify the hundreds of billions being poured into building better models.
What's On The Pod
Some new podcast episodes
The Cognitive Revolution — AI:AM Highlights: Clean Neurons, Black October, the GPU Credit Crunch, and Buying the Supply Chain