Ad
Skip to content

Matthias Bastian

Matthias is the co-founder and publisher of THE DECODER, exploring how AI is fundamentally changing the relationship between humans and computers.
Read full article about: Google relaunches its AI creative studio Flow with new features and integrations

Google has relaunched and expanded its AI creative studio Flow. The company's image generation experiments, Whisk and ImageFX, are now being integrated directly into Flow, and starting in March, users will be able to transfer their existing projects and files. At the core is Google's image model Nano Banana, which lets users generate images and use them directly as the basis for videos with Veo.

Other new features include a lasso tool for targeted editing of image areas using text input, flexible media management with collections, and tools for extending clips and controlling camera movements. Google is aiming to combine text, image, and video creation into a single workflow.

Flow is available at flow.google and free to use after signing up - paying users get higher usage limits and access to the full set of tools. According to Google, users have created over 1.5 billion images and videos since the platform launched last year.

Read full article about: Adobe's new Firefly "Quick Cut" tool turns raw footage into a rough edit from a text prompt

Adobe has added a new feature called "Quick Cut" to its Firefly AI creative platform. The tool lets video creators upload their own raw footage or generate new material with AI, then automatically produces an initial rough cut. Users describe what the video should be about in plain language—an interview, a product demo, a travel vlog—and Firefly builds a structured first edit from that description. Scripts or shot lists can also be added as optional input.

Quick Cut targets product reviewers, reporters, podcasters, and marketers. Firefly bundles AI models from Adobe, Google, OpenAI, and Runway into a single app. Through March 16, Adobe is offering unlimited image and video generation in up to 2K resolution on select subscription plans.

Read full article about: Anthropic refuses Pentagon demand to loosen military AI restrictions, faces Defense Production Act threat

Anthropic won't back down on its military AI restrictions, but the Pentagon is giving it little choice. According to Reuters, the AI company continues to refuse to loosen its safety guardrails for military use. The dispute centers on security measures that prevent Anthropic's technology from being used for autonomous weapon control and domestic surveillance.

At a meeting between Anthropic CEO Dario Amodei and US Secretary of Defense Pete Hegseth, Hegseth delivered an ultimatum: either Anthropic complies by Friday, or the Pentagon will invoke the Defense Production Act—a law that can force companies to cooperate—or classify Anthropic as a supply chain risk. According to Franklin Turner, a government contracts attorney at McCarter & English, such a move against Anthropic would be unprecedented and could trigger a wave of lawsuits.

Amodei argued that the existing safeguards don't interfere with current military operations. Meanwhile, the Pentagon is negotiating parallel AI contracts with Google, xAI, and OpenAI for battlefield applications, including autonomous drone swarms, robots, and cyberattacks. Elon Musk's xAI has already secured an agreement with the Pentagon to deploy on classified networks this week.

Read full article about: OpenAI ships API upgrades targeting voice reliability and agent speed for developers

OpenAI has shipped two API updates for developers: the new gpt-realtime-1.5 model for the real-time API is designed to make voice commands more reliable. In internal testing, OpenAI saw roughly a ten percent improvement in transcribing numbers and letters, a five percent bump in logical audio tasks, and seven percent better instruction following. The audio model has also been updated to version 1.5.

The Responses API also now supports WebSockets. Instead of retransmitting the full context with every request, this opens a persistent connection that only sends new data as it comes in. According to OpenAI, the change speeds up complex AI agents with many tool calls by 20 to 40 percent.

Read full article about: Google, OpenAI, and Anthropic are all bracing for Deepseek's next big release

Chinese AI startup Deepseek has apparently trained its latest AI model on Nvidia's most powerful Blackwell chips, despite the US export ban. That's according to Reuters, citing a senior Trump administration official. The model is expected to drop next week. Rumors about chip smuggling had already been circulating since late last year.

The official says the Blackwell chips are believed to be in a data center in Inner Mongolia, and Deepseek is expected to scrub technical fingerprints of US chip usage before release. The official wouldn't say how Deepseek obtained the chips. Nvidia declined to comment, and neither Deepseek nor the US Department of Commerce responded to Reuters.

If the timing of these leaks is any indicator, Deepseek may be on the verge of another major splash. Google, OpenAI, and Anthropic have all been complaining about distillation attacks on their models by Chinese startups, and OpenAI recently moved to relativize a well-known coding benchmark. Together, these moves suggest Deepseek is about to deliver strong results at rock-bottom prices once again. Back in January 2025, China's leading AI startup sent shockwaves through US tech stocks riding the AI bubble.

Read full article about: Anthropic accuses Deepseek, Moonshot, and MiniMax of stealing Claude's AI data through 16 million queries

Anthropic says it has caught Chinese AI labs Deepseek, Moonshot, and MiniMax running large-scale distillation attacks on Claude, a technique where a weaker model learns from the output of a stronger one. Over 24,000 fake accounts fired off more than 16 million queries targeting Claude's reasoning, programming, and tool usage capabilities. The labs used proxy services to bypass China's access restrictions.

Lab Requests Targets
Deepseek 150,000+ Extracting reasoning steps, reward model data for reinforcement learning, censorship-compliant answers on politically sensitive topics
Moonshot AI 3.4 million+ Agent-based reasoning, tool usage, programming, data analysis, computer vision, reconstructing Claude's thought processes
MiniMax 13 million+ Agent-based programming, tool usage and orchestration; pivoted to new Claude model within 24 hours

Deepseek specifically targeted Claude's reasoning chain, extracting thought processes and censorship-compliant answers on sensitive topics. MiniMax ran the biggest campaign by far with over 13 million requests. When Anthropic shipped a new model, MiniMax pivoted within 24 hours and redirected nearly half its traffic to the updated system, Anthropic says.

OpenAI and Google report similar attempts from Chinese labs. Anthropic is calling on the industry and policymakers to mount a coordinated response.

Read full article about: OpenAI wants to retire the AI coding benchmark that everyone has been competing on

OpenAI says the SWE-bench Verified programming benchmark has lost its value as a meaningful measure of AI coding ability. The company points to two main problems: at least 59.4 percent of the benchmark's tasks are flawed, rejecting correct solutions because they enforce specific implementation details or check functions not described in the task.

Many tasks and solutions have also leaked into leading models' training data. OpenAI reports that GPT-5.2, Claude Opus 4.5, and Gemini 3 Flash Preview could reproduce some original fixes from memory, meaning benchmark progress increasingly reflects what a model has seen, not how well it codes. OpenAI recommends SWE-bench Pro instead and is building its own non-public tests.

There's a possible strategic angle here: a "contaminated" benchmark can make rivals—especially open-source models—look better and skew rankings. SWE-bench Verified was long the gold standard for AI coding evaluation, with OpenAI, Anthropic, Google, and many Chinese open-weight models competing for small leads. AI benchmarks can provide useful signal, but their real-world value remains limited.