OpenAI Claims AI Refuted the Exponential Time Hypothesis
ALSO: Walmart welcomes AI shoppers, AI security jobs up 268%
Welcome to Guru99 AI Report!
Top Story: OpenAI claims its AI just broke a famous computing rule, and top experts are now lining up to check it. Plus, why Walmart welcomes AI shoppers while Amazon shuts them out, and which cyber role jumped 268%.
🧮 OpenAI Says Its AI Toppled a Famous Computing Rule
Brief Buzz:
OpenAI made two major announcements this morning. One is that its AI has solved some long-standing open problems in mathematics, computer science and physics. The other is a much cheaper model that can run your apps for you. Both claims still need to be verified by others.
- OpenAI claims its model has refuted the Exponential Time Hypothesis (a key assumption that logic puzzles can’t be solved much faster).
- Two of the hypothesis’s original authors (Russell Impagliazzo and Ramamohan Paturi) will be leading the independent review, with MIT’s Ryan Williams.
- Nobel laureate David Gross is leading a separate group looking at OpenAI’s claimed physics proof (which is not yet public).
- GPT-6.1 Terra costs 1/5th the price of GPT-6.1 Sol per million input/output tokens and can run apps on screen.
- OpenAI says Terra is over 3x faster than its more expensive models, but independent tests put it below Sol on coding.
💡 Why Should You Care?
If OpenAI is correct, experts may have to go back and rethink how fast computers can solve hard problems. Terra matters sooner.
🛒 Walmart Welcomes AI Shoppers. Amazon Shows Them the Door
Brief Buzz:
AI agents can now shop for you. Retailers, however, want a say in which ones get the green light. Walmart is opening its doors to third-party agents (with appropriate guardrails), whereas Amazon is keeping them out and going the route of its own assistant.
- Walmart’s Agent Profiles are launching in November. Each AI agent will have its own profile, its own purchase history and an optional spending limit.
- Approval is the default. Walmart can ask for further approval even when an agent is within budget, and agents never see your card details.
- Amazon has blocked Meta’s Muse, is taking legal action against Perplexity’s Comet and is promoting its own Alexa for Shopping.
- Both Meta’s Muse and Anthropic’s Claude Duet require your approval before purchase, and both use one-off virtual cards.
- Six banks (including Bank of America and NatWest) have warned that agents may enter card details on websites or may favor weaker protections.
💡 Why Should You Care?
For shoppers: keep approval on. For testers: agent checkout is a new surface. Approval flows, spending-limit edge cases and agents going off script are all things you’ll need to watch.
🤖 The Job-Hunt AI Catch-22: Use It, Don’t Sound Like It
Brief Buzz:
Job seekers are in an AI catch-22. They use AI to find work, then go through their applications to make sure they don’t sound like a bot wrote them. That’s what Greenhouse (the hiring software company) found in a survey of over 2,800 job seekers across five countries. The figures below are US only.
- 77% used AI in their search over the past year, yet 60% have rewritten or are considering rewriting their applications to avoid the AI-written feel.
- 81% were rejected without a single conversation in the last year (not even with an AI).
- 62% attempted to game automated screening; of that group, 40% said they didn’t actually have the skills, and a third planted hidden text for AI reviewers.
- Only 7% would flatly refuse an AI interview, with 49% wanting a human to review their answers. (Greenhouse has AI interview tools.)
- AI agents: 69% would be OK using one, but just 5% would be happy to have one apply for jobs on their behalf.
💡 Why Should You Care?
Job hunting? The game now runs both ways: write for the bot, then prove you’re human. Hiring? 88% of candidates want legal rules requiring transparency about AI in interviews, and the first step is giving them that notice upfront.
🧬 Anthropic Puts Claude to Work Inventing Actual Medicines
Brief Buzz:
Anthropic has just published its first progress report on drug development. Its science team spent months embedded in six real drug programs, and it has also bought a one-person biotech. The results are early, preclinical and largely self-reported.
- 35,000 antibody designs in just weeks at Divergence Labs; 20 dislodged IgE, and three outperformed Xolair in the lab.
- Generate Biomedicines shrank a CAR component down to 45 amino acids (under a third of the clinical standard), boosting receptor counts up to 13x.
- At Enveda, Claude went through 1,061 pages of manufacturing records, cutting a 26-hour manual check to under 3 hours.
- Anthropic acquired Mithril Medicine and its rare muscular dystrophy program; one of its antibodies roughly doubled forelimb force in a small mouse pilot.
- All of it is preclinical (no human trials), and the partners graded their own results.
💡 Why Should You Care?
If it holds up, AI goes from writing code to actually running real drug programs, faster and cheaper, for diseases big pharma won’t touch. The catch: early mouse data rarely makes it to human trials.
🎧 Google’s Genie 4 Builds Walkable AI Worlds That Talk Back
Brief Buzz:
Most AI video tools produce a clip you just watch. Google DeepMind’s new Genie 4 is different. It’s a “world model”: an AI that creates a world you can walk around in and modify as you go. It’s DeepMind’s first real-time model with native sound and music.
- Move with keyboard controls and reconfigure scenes using plain-English prompts (e.g. change the weather, add a person).
- Simulated people (which DeepMind calls its biggest leap) move naturally and talk in real time, responding to you and to the world around them.
- Gemini models steer between your edits, while Gemini 3.8 Flash checks the frames to ensure each planned step actually worked.
- In a robotics demo, Gemini 4 Argon and Genie 4 were shown simulating Apptronik’s Apollo humanoid playing checkers.
- It’s a research preview for trusted testers. Google AI Ultra subscribers and early-access Gemini API partners come next (no date given).
💡 Why Should You Care?
Google is positioning Genie 4 as a practice space for robots, and testers are already using it to rehearse film scenes and prototype games. The caveat: details still drift during longer sessions.
📉 US Cyber Job Postings Fall 26%; AI Security Bucks Trend
Brief Buzz:
US cyber jobs dropped significantly this year. CyberSN’s latest jobs report shows 235,786 postings from January to September 2026, 26.4% fewer than the same period last year. Several AI-related roles, however, went the other way.
- Compared to 2024, postings are down just 3.7%. 2025 was the best year in CyberSN’s four years of data.
- 9 of the 10 job functions and 36 of the 41 tracked roles had fewer postings than the previous year.
- AI Security Engineer postings jumped 268% to 2,409 (though the series overlaps with general Security Engineer postings).
- Cyber Data Scientist posts tripled to 1,493 from a very small base, with just over half coming in September alone.
- 14.6% of postings referred to AI or machine learning, much the same as 15.1% the year before.
💡 Why Should You Care?
These are job ads, not hires, and CyberSN doesn’t say why postings declined. For security professionals, the clear growth areas are AI security and data science.
Hey! I’m Krishna Rungta
Founder of Guru99.com, Editor-in-chief & Technology Expert
Was this email forwarded to you? Sign up for free here.
