The AI Times
EN
Volume I · Daily editionDaily roundup of AI newsMCPMD
EN
THE AI TIMES
Apps · Privacy · AI capabilities

AI moves closer to everyday tasks

Seznam is preparing to add AI to more services, and Groupon is opening experience searches to programs that work on behalf of customers. Both stories hinge on the difference between plans and what already works. Codex, meanwhile, is changing the availability of an older model. The research findings below show further AI capabilities, but also their limits: success on selected physics problems does not guarantee an error-free answer, and response speed can decide the outcome even against a stronger model.

Seznam · Apps16 October

Seznam wants to bring its AI to maps and horoscopes

Seznam plans to expand its artificial intelligence to more products. According to Živě.cz, it already responds twice as fast as three months ago.

Seznam AI · service statusŽivě.cz · publication
Seznam AISeLLMa
response speed · 2x faster
comparison · three months ago
maps · planned expansion
horoscopes · planned expansion
In-⁠house technologyPlans for more products

For people using Seznam services, the key development is the plan to add AI to maps and horoscopes too.

Živě.cz reports on the planned expansion; this does not confirm that all the new features are already working.

The service is built on SeLLMa technology developed in-house. Živě.cz reports weekly user numbers in the high hundreds of thousands. According to the article, they conduct conversations in the low millions.

SOURCEzive.cz
Shopping · AI16 October

AI will find an experience for you. Groupon opens up its listings

The new LivingSocial platform lets AI search for and compare experiences and prepare a purchase.

According to Lupa.cz, Groupon has opened up to purchases mediated by AI agents, programs that carry out assigned tasks. The article does not demonstrate automatic payment without human involvement.

SOURCElupa.cz
Codex · Models14 October

Codex retires GPT-5.5 for ChatGPT sign-⁠ins

OpenAI recommends moving to GPT-6.1 Sol where the subscription and application in use make it available.

The Codex changelog states that GPT-5.5 was retired from this service on 14 October at 19:00 UTC for users signed in through ChatGPT.

Rules · Security15 October

FTC reportedly investigates OpenAI and Anthropic security

According to an Axios source, the US agency is examining whether AI security incidents harmed consumers.

For users, the key question is whether AI service offerings match their actual security. Ars Technica AI reports that the US Federal Trade Commission is investigating OpenAI and Anthropic. The report relies on an Axios source; Ars was unable to review the agency’s requests itself.

This is not a published finding that the companies harmed consumers. According to the article, the agency sent out investigative requests for information on Wednesday.

Security · Research14 October

Small AI could spread malware. For now, it is a warning

Ars Technica describes Anthropic’s warning about AI spreading between smart devices in the future. The experiment also revealed major obstacles.

Ars Technica AI reports that smaller models could eventually help malware spread independently between internet-connected devices. This is not a documented attack on households; the experiment revealed significant difficulties. The text lists smart fridges and vacuum cleaners as examples of devices at risk. Unlike conventional malware, AI could explore unfamiliar devices on its own.

Science · AI14 October

AI designs enzymes that cut selected proteins

Researchers used AI to design proteins that cleave other proteins. They also demonstrated effects on targets linked to neurological diseases.

For general readers, this is research news, not an announcement of an available treatment. According to Ars Technica AI, scientists used AI tools to redesign enzymes called proteases, which break down proteins.

The design allows cleavage to target a specific protein. The researchers started with the catalytic core of two types of proteases. They replaced the rest of their protein structure with a new design. Bonds between amino acids are inherently very stable.

Models · Privacy15 October

NVIDIA adds models for AI on your own computer

The new Bolt models also target local use. The larger version is expected to fit in 24 GB of memory; the smaller one arrives on 23 October.

For users, the key benefit is the ability to process some tasks directly on their own device. NVIDIA AI Blog announced the release of Nemotron 3.5 Super VL and Bolt 74B-A9B; the smaller Bolt E4B is promised for 23 October. These are models to embed in tools, rather than a new ready-made application for everyone. According to the manufacturer, Super VL can work with text, tables and charts on web pages and in PDFs.

What experts say

Simon Willison

developer, creator of the LLM and Datasette tools

Chirp Pong

Willison is testing a two-player game on two nearby phones that exchange state through sounds picked up by their microphones. He says the experiment works even in static HTML. He asked Claude Opus 5.5 Extra in Claude Code to build the game; it is a concrete example of creating an application with AI.

simonwillison.net
What experts say

Arvind Narayanan a Sayash Kapoor

Princeton researchers, authors of AI as Normal Technology

Steering AI Beyond Automation

The article rejects the idea that replacing human labor must be AI’s sole purpose. It proposes systems that show people sources, assumptions and work in progress, and let them intervene throughout. It also considers fast responses an important condition for collaboration, so that waiting for AI does not break concentration.

normaltech.ai
Science · AI capabilities15 October

GPT-6-Astra solves 49 of 50 physics problems in a test

The Decoder describes results on previously unpublished theoretical physics calculations. Success in the test does not make it an independent scientist.

TheoPhysEval · problems and cost per problemThe Decoder · publication
Physics problems solved49 of 50 problems0 problems50 problemsCost per physics problem2.40 USDGPT-6-AstraPreviously unpublished calculationsResearch test

For users, this demonstrates AI’s capabilities in demanding calculations, rather than guaranteeing that every answer is correct. According to The Decoder, GPT-6-Astra completed 49 of 50 problems in the TheoPhysEval test with accuracy comparable to leading experts. The article’s headline gives a cost of USD 2.40 per problem; this figure should not be confused with a subscription price.

Research · Speed16 October

When the world will not wait, slower AI can lose

A new test keeps games running while AI thinks. The results show that the right answer can arrive too late.

For users, the key point is that response speed can sometimes matter as much as reasoning quality. A paper published on arXiv cs.AI introduces RealtimeBench, with 180 tasks across six types of games. GPT-6.1-sol beats the faster Jev when the game is paused, but the ranking reverses when it runs continuously. This is a research test, not an overall ranking of model quality.

SOURCEarxiv.org
Development · Security13 October

GitHub deploys smaller AI to find leaked secrets

The new check combines known-pattern detection with a model that assesses the surrounding text. GitHub says it reduces false alarms.

This protects credentials accidentally stored in code. GitHub Copilot Changelog describes a model that distinguishes real secrets from similar-looking examples. The results are vendor measurements, not an independent security audit of all projects. Running the classifier cost more than 98% less than remotely calling the previous model.

Development · Practice15 October

Netflix improves AI code navigation with a JetBrains tool

It gave AI assistants an accurate map of relationships within a project. In tests, this typically reduced the amount of text processed and the time needed.

This is about more efficient application development, not a new Netflix feature. According to JetBrains AI blog, Netflix connected Claude Code and Codex to a JetBrains tool that understands project structure. Measurements on two real projects showed typical efficiency gains of 20 to 30%. Here, efficiency combines savings in time and processed text. The model, task and project stayed the same in the comparison; access to the JetBrains tool changed. Result correctness did not deteriorate in the runs described. Netflix plans similar measurements for Python, TypeScript/JavaScript and Go.

VerifiedMyth of the day

“A photograph shared on X shows a woman wearing a Release the Epstein files T-shirt being punched at a Trump rally.”
False

Lead Stories identified the shared image as AI-generated based on two detection tools and a comparison with news photographs. The actual incident at the rally in Youngstown involved a different woman with different facial features and clothing.

Verified by: leadstories.com

Repo of the dayGitHub Trending

MCPClaude CodeReverse engineering

morluto/rea

REA: Reverse Engineer Anything

A single MCP interface through which an AI agent (Claude Code, Codex, Cursor or Gemini CLI) explores an application without source code: native binaries, JavaScript and Electron applications, .NET and websites. It adds evidence and limitations to each conclusion, and the analysis runs on your own computer.

Total stars
68 211
Forks
14 360
Language
TypeScript
What it does
  • Explains how a feature in another application works, shows evidence and builds a similar feature into your project.
  • Connects to Ghidra, Hopper or IDA to analyze native code.
  • Analyzes JavaScript and Electron without an additional tool; the same features are also available in the terminal.
Get started
npx rea-agents setup

Repository on GitHub →

Tip to tryTip of the day

Development with AI

Try checking a rule before committing code changes

Difficulty
Advanced
Time
20 minutes

Based on: simonwillison.net

  1. Set up a pre-commit check that rejects direct calls to diagnostic modules that bypass the required library.
  2. Run it on a valid code change and verify that it accepts it.
  3. Run it on a change that bypasses the library and verify that it rejects it.

Events in CzechiaNext 14 days · until 29 October

  1. Rakathon 2026

    Help change the future of cancer care with AI!

    Česká asociace AI

    Prahaprice on website
  2. AI in culture: Why and at what cost?

    A conference on practical uses of artificial intelligence in museums and cultural institutions.

    ConferencePrahaprice on website
  3. AI Monday Praha #16

    Join an open community of people who actually use AI.

    MeetupPrahafree
  4. Experience sharing for CMOs: AI in B2B marketing – how to fill your pipeline with quality leads effectively

    MeetingBrnofree
  5. Automation & AI Clinic

    An event for companies that want to discuss a specific automation, digitalization or AI problem directly with specialists in small expert zones.

    WorkshopPrahafree
  6. ConferencePrahaprice on website
  7. AI Restart 2026

    A full-day conference packed with visionary perspectives and practical demonstrations of how to use artificial intelligence in your work.

    Česká asociace AI

    ConferencePrahaCZK 4,290
  8. AI conference for lawyers

    Whether you are an in-house lawyer or handle HR, this is the only conference in the Czech Republic that shows the real intersection of AI and law.

    Česká asociace AI

    Kino Přítomnostprice on website

Model leaderboardIntelligence Index · top 10

Which models are the smartest today

Artificial Analysis combines several independent benchmarks into one number. A higher score means the model handles harder tasks.

  1. 58Claude Opus 5.5
  2. 56Claude Sonnet 5.5
  3. 56Muse Wonder Full
  4. 53Claude Fable 5.1
  5. 53GPT-⁠6 Astra
  6. 53Gemini 4 Argon
  7. 52Muse Wonder
  8. 52GPT-⁠6.1 Sol
  9. 51Claude Opus 5
  10. 50Claude Fable 5

Cost per completed task

  1. Claude Opus 5.55,98 $
  2. Claude Sonnet 5.55,46 $
  3. Claude Fable 5.17,63 $
  4. GPT-⁠6 Astra3,26 $
  5. Gemini 4 Argon1,99 $
  6. Muse Wonder4,52 $
  7. GPT-⁠6.1 Sol0,72 $
  8. Grok 4.73,74 $
  9. MiMo-⁠V2.6-⁠Pro0,13 $
  10. Qwen3.8 Max5,41 $

Best image models

  1. GPT Image 2.5 Sunburst (max)1218
  2. GPT Image 2.5 Flare (max)1207
  3. GPT Image 2 (high)1185
  4. Nano Banana 2.11172
  5. Grok Imagine Image 2.01165
  6. MAI-⁠Image-⁠2.61165
  7. Nano Banana 2 (Gemini 3.1 Flash Image)1139
  8. Muse Image1127
  9. GPT Image 1.5 (high)1122
  10. MAI-⁠Image-⁠2.6-⁠Flash1117

Data: Artificial Analysis · Intelligence Index, higher is better, cost per completed task (Intelligence Index task average) and Text to Image Arena · as of Oct 16, 2026