Author HubMatthias Bastian

Matthias is the co-founder and publisher of THE DECODER, exploring how AI is fundamentally changing the relationship between humans and computers.
AI research
Update

So-called reasoning models are more efficient but not more capable than regular LLMs, study finds

AI in practice

Google leans on token metrics, not revenue, adding to bubble talk about AI growth

AI and society

The long-predicted deepfake dystopia has arrived with Sora 2

Ad
AI in practice
Short

Microsoft is adding new AI shopping tools to its Edge browser in the US. The built-in Copilot can now surface price comparisons, price histories, and cashback options right inside the browser. Users can open the feature from the sidebar icon, which then pulls up product details and sends price alerts automatically. A new "Copilot Mode" also flags cheaper deals or active discounts when they appear. All features are optional and currently limited to the U.S.

OpenAI recently moved in a similar direction with ChatGPT Shopping, its new e-commerce interface focused on personalized product research.

AI in practice
Short

Meta is kicking competing AI chatbots off WhatsApp. OpenAI and Microsoft have announced that ChatGPT and Copilot will be forced to leave the messaging app due to changes in the WhatsApp Business Solution terms of service, which go into effect on January 15, 2026. Under the new rules, Meta prohibits AI companies from using the Business API as a platform for their own chatbots. The ban specifically targets cases where the AI itself is the product, meaning customer service and support chatbots from other companies are still allowed.

ChatGPT and Copilot will continue to work in WhatsApp until the deadline hits. ChatGPT users can link their accounts to preserve their chat history, but Copilot users won't have that option. It's likely that other providers, such as Perplexity, will also have to leave WhatsApp soon. Once the ban takes effect in January, Meta AI—powered by the currently lagging Llama models—will be the only general-purpose AI chatbot left standing in WhatsApp.

Ad
Ad
AI in practice
Short

Claude Opus 4.5 scores higher than its rivals in prompt-injection security, but the results show how limited these defenses still are. A benchmark by the security firm Gray Swan found that a single "very strong" prompt injection attack breaks through Opus 4.5's safeguards 4.7 percent of the time. Give an attacker ten attempts and the success rate jumps to 33.6 percent. At 100 attempts, it reaches 63 percent. Even with those gaps, Opus 4.5 still performs better than models like Google's Gemini 3 Pro and GPT-5.1, which show attack rates as high as 92 percent.

Anthropic

Prompt injection works by slipping hidden instructions into a prompt to bypass safety filters, a long-standing weakness in large language models. The issue becomes even more serious in agent-style systems, which expose more potential entry points and make these attacks easier to exploit.

Ad
Ad
AI in practice
Short

Amazon has announced a major investment in its AI footprint for federal work, saying it will spend up to $50 billion to expand AI and supercomputing infrastructure for U.S. government agencies. The project supports the U.S. government’s AI Action Plan and is expected to help agencies accelerate discovery, decision-making, and mission workflows, including through faster analysis and automation.

Amazon’s investment underscores the strategic importance of AI and supercomputing in maintaining technological superiority, safeguarding critical infrastructure, and driving industrial innovation.

Starting in 2026, AWS plans to add nearly 1.3 gigawatts of new compute capacity across its Top Secret, Secret, and GovCloud (US) regions. Once live, agencies will be able to use services such as SageMaker, Bedrock, Trainium chips, and Anthropic models to build their own AI applications, speeding up data analysis and improving workflows in areas like cybersecurity, healthcare research, and autonomous systems.

Ad
Ad
AI research
Short

Google has added a slide generator to NotebookLM, giving users a quick way to turn their sources into simple slide decks. The tool can help structure notes or produce early drafts, and Google says it can also enhance existing slides visually.

Right now, NotebookLM delivers slides only as PDFs. Export options for Google Slides and PowerPoint are in development, Google says. The feature is available immediately, with daily usage limits based on the user's account.

The slide tool, along with a new infographic feature, runs on Google's Nano Bana Pro model (Gemini 3 Pro Image Generation). It is the first model capable of turning highly detailed prompts into precise, text-heavy images.

Google News

Bank details

IBAN: DE88 2507 0070 0053 0014 00
BIC: DEUTDE2HXXX
Account holder: Deep Content GmbH
Purpose: Support THE DECODER
AI research

The ARC benchmark's fall marks another casualty of relentless AI optimization

AI in practice

DeepseekMath-V2 is Deepseek's latest attempt to pop the US AI bubble

AI research

Frustrated authors withdraw papers after realizing their reviewers are just lazy LLMs

Google News