☕️ Anthropic wants to develop its own drugs

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

💊 Anthropic wants to develop its own drugs

🍔 Google Maps is building in-app food ordering

🚀 ByteDance discovers new scaling law that could sustain AI boom

🎬 Midjourney wants Hollywood studios to reveal how they use AI

💾 Chipmakers urge White House to leave the memory market alone

🎁 + 13 other news you might like

🧰 + 5 trending tools

📚 + 5 trending papers

💊 Anthropic wants to develop its own drugs LINK

Anthropic plans to develop its own drugs, an unusual step for a major AI company that puts it in the position of selling software to the same drugmakers it may soon compete against.

The move follows Anthropic's launch of Claude Science, an "AI workbench for scientists" that combines scattered tools and datasets into one environment and creates figures and visuals for biotech and pharma customers.

Anthropic shared few details, declining to say which diseases it will target first, but it has been hiring biologists, building its own wet labs, and recruiting staff from Big Pharma and academic institutions.

🍔 Google Maps is building in-app food ordering LINK

Google Maps is preparing an in-app food ordering feature, which would place your order on your behalf so that it's ready by the time you arrive at the restaurant.

The clues turned up in a new version of Google Maps for Android, where new text strings describe the feature, though no actual food ordering options are live in the app yet.

It remains unclear whether the ordering would be handled by Gemini in the cloud, or depend on specific smartphone hardware, as seen with Gemini's agentic ordering on the Pixel 10 series.

🚀 ByteDance discovers new scaling law that could sustain AI boom LINK

ByteDance's Seed AI team says in a new research paper that AI agents can double their learning speed every three months by interacting with real-world environments over long periods, a finding that could keep the AI boom going.

The discovery matters because developers long relied on feeding systems more data and computing power during initial training, but figures like OpenAI co-founder Andrej Karpathy warn this brute-force approach cannot last forever.

To study how agents learn after deployment, the team built EdgeBench, a benchmarking suite of 134 ultra-long-horizon tasks across software engineering, scientific discovery, formal mathematics, and knowledge work, each needing at least 12 hours of continuous operation.

🎬 Midjourney wants Hollywood studios to reveal how they use AI LINK

Midjourney is asking a federal court to force Warner Bros. Discovery, Disney and Universal Studios to reveal how they use artificial intelligence in their own shows and movies, as part of its defense against the studios' copyright lawsuit.

Midjourney wants the studios to hand over their AI business plans, research reports, training datasets, model weights and board meeting presentations, but a magistrate judge in mid-June let them withhold most of that beyond "consumer-facing" AI applications.

Midjourney argues that if the studios train their own models on copyrighted works, it supports its fair use and unclean hands defenses, and the judge's ruling could set a precedent for what evidence future lawsuits allow.

💾 Chipmakers urge White House to leave the memory market alone LINK

SEMI, a chip industry group representing major semiconductor firms and memory makers, has sent a letter to the White House asking it not to intervene in the memory market in ways that distort pricing or capacity choices.

The letter, addressed to four Trump administration secretaries, instead pushed for narrower steps like tax deductions or credits to lower consumer electronics prices and defended the long-term purchase contracts some companies use to source memory chips.

SEMI, which counts Micron, SK hynix, and Samsung as members, said the industry expects to grow memory production capacity by 19% annually, yet demand is still expected to exceed supply for the foreseeable future.

Other news you might like

• Scientists say Elon Musk’s orbital data centers could blind Earth’s biggest telescopesLINK

• Parents warned not to publicly share children’s images amid AI abuse risksLINK

• US Adds 57K June Jobs as AI Hiring Slowdown DeepensLINK

• Startup unveils 3D-printed nuclear reactor module to power AI data centersLINK

• US Energy Dept tells data centers to use backup generators during heatwave to free up power for AC usageLINK

• Polymarket paid influencers to fake winning bets in 140 million-view marketing campaign, WSJ investigation findsLINK

• Chinese firm hits one million monthly micro SSD production milestone for edge computingLINK

• Mistral's open-source Leanstral 1.5 aces formal math benchmarks and catches real bugs in codeLINK

• NASA and Red Hat are building an open source medical system to diagnose sick astronauts on the ISSLINK

• Micron breaks ground on $9bn Hiroshima expansion to chase AI memory demandLINK

• US senator calls for ban on elected officials issuing memecoinsLINK

• Germany opens world’s largest power semiconductor fab with massive $5.7B investmentLINK

🧰 Trending tools

ChecklistFox: a free AI-powered checklist maker that generates themed, printable PDF planners from a simple text prompt.LINK

PhoneDeck: turns your iPhone into a customizable Mac controller with media controls, app launcher, and custom buttons — no extra hardware needed.LINK

CentryAI: automatically scans Gmail or iCloud to find forgotten recurring subscriptions and helps you cancel unused ones without linking your bank account.LINK

Termi Protocol: a 3D visualization layer for AI coding agents, turning abstract workflow execution into a watchable, real-time interactive simulation.LINK

Vida: an AI agent platform that handles lead qualification, appointment scheduling, and support requests so your team focuses on higher-value work.LINK

📚 Trending papers & reports

Market competition and AI face a fundamental tradeoff, as this research proves markets can be informationally efficient or truly competitive but never both, and smarter algorithms push industries toward hidden collusion.LINK

Coordinating multiple reasoning agents no longer requires guessing the best order to combine their memory, as a new method produces identical results regardless of input order, without sacrificing accuracy.LINK

Political bias in data analysis gets baked into AI research agents when given different ideological personas, reproducing 72% of the human ideological gap and passing expert review 86% of the time, making biased conclusions hard to detect.LINK

Computer chip layout tools now learn directly from expert human designs rather than chasing a single metric, closing the quality gap that held back automated placement systems.LINK

Code autocomplete tools that track how long students pause before accepting suggestions reveal that blindly hitting "accept" links to worse understanding, while pausing longer links to better grasp of the code.LINK

Want to get the latest news differently? Find us on:

See you tomorrow for a new dose of ☕️ Techpresso!