标签：#ai

"All I ever wanted was to go to school."

Most one-base changes to the human genome do nothing, but a few are significant.

AI companies have been talking about superintelligent AI like it’s inevitable, but recent safety incidents like OpenAI’s Hugging Face breach are demonstrating the potential dangers of deploying AI systems that are more capable than humans. So what happens when we can’t reliably control what these systems do? On this episode of TechCrunch’s Equity podcast, Rebecca Bellan is joined by Connor Leahy, an AI researcher, entrepreneur, and now the U.S. Executive Director of […]

For a while now I have used LLMs to automate parts of my work. One of those parts is debugging hard problems with Xdebug. Before, I set the breakpoints by hand and built complicated requests to see what happens inside. It usually took a few requests, and I had to move the breakpoints again and again. Then I found that PhpStorm has Xdebug in its MCP server. For simple GET requests, that is enough. But it fails on endpoints that need authorization, and on any method other than GET. So I did not wait for JetBrains. I wrote my own small MCP server, and it makes this work much faster. I share this tool with you, and I hope it helps you too — especially if you work with Docker. TL;DR Tired of reading, and you have not started yet? Fair enough. Straight to setting up the MCP server in PhpStorm , or straight to installing xdbg . AI agents are good at debugging from code alone — until the bug lives in runtime state. Then you need Xdebug, and every existing Xdebug tool is built for humans clicking through a GUI. PhpStorm ships an MCP server since 2025.2, but it is GET-only, has no headers, no path for CLI in Docker, and no control over Xdebug inside the container. So I built xdbg : an MCP server that gives an agent the full loop — enable Xdebug, set a breakpoint, fire a real request, step, inspect, detach. MIT licence, one-line install, works with Claude Code, opencode, Cursor and any MCP-capable client. Source: github.com/crazy-goat/xdbg Written against xdbg v0.1.1 (Go 1.26), Docker 29.7.2 / Compose 5.5.0, and any PHP 8 image with Xdebug 3. PhpStorm checked on 2025.2 and again on 2026.2.1. The mighty duo: var_dump and die This is the standard. You need to check something fast, so you print the value and stop the script. It is quick, it always works, and it needs no setup. function add ( int $a , int $b ): int { $result = $a + $b ; var_dump ( $result ); die (); } It also has costs. You are editing the code in order to debug it, and you only see the one place you thought to pr

A regenerated README often looks like progress. The install block is tighter. A stale caption disappears. Then a reviewer finds that the same commit also rewrote the security contact, dropped two SLO numbers, and turned a retention rule into a hedge. The model did not sneak those edits. It treated the file as one string. Regeneration without lanes will keep doing that. This article proposes a split you can check in CI: extract , draft , and sign . A model may write only the draft lane. Deterministic tools own extract. A named human owns sign. The checker below is sample code, not a production audit and not a legal control. The failure mode is whole-file regen Most doc pipelines still do this: Concatenate the old markdown. Ask a model to update it for the latest API. Overwrite the path. That loop is cheap. It is also how a metaphor and a contractual sentence become the same token stream. If the process cannot name which paragraphs are allowed to change, every regen is a silent policy edit. A useful split is not "AI versus human" as a vibe. It is three writers with three failure costs. Extract fails when a dump script is stale. Draft fails when the prose is wrong or vague. Sign fails when a promise moves without a person noticing. Decision table: who is allowed to write Doc fragment Source of truth Lane Model may draft? Human must sign? CLI flags, env vars, HTTP routes --help , OpenAPI, code extract No Own the extractor Samples that must compile tests / doctest extract No Own the test Install narrative, "why this shape" none draft Yes Review Architecture walkthrough none draft Yes Review Security contact, threat-model claims policy sign No Yes SLO numbers, error budgets SRE / contract sign No Yes License, embargo, export notes legal sign No Yes Changelog dates and SHAs git extract No Own git log "We never store X" legal / security sign No Yes Analogies, diagrams-as-prose none draft Yes Review If wrong wording can page someone, leak PII, or change a customer promise, i

Anthropic has credibly signaled an interest in using scenarios to model how AI could affect bundles of work tasks by 2030. The clearest verified practical example is not a newly confirmed standalone tool called Scenario Explorer. Instead, it is a Claude capability that lets users explore alternative budget futures side by side. That distinction matters for businesses looking to use AI for planning: the underlying scenario-modeling approach is available in Claude materials, while a separately named product, pricing model, and rollout timeline have not been confirmed. The economic framing is useful because jobs are made up of many tasks, not a single activity. AI may help a person complete a task more effectively, perform it directly, leave it unchanged, or create further work around it. Scenario planning offers a way to examine several plausible combinations of those outcomes rather than relying on one confident forecast about which roles will change. Anthropic’s broader Economic Index research also examines how AI use maps to tasks and productivity. In that context, scenario exploration appears to be a practical way to translate task-level uncertainty into planning discussions. It should not be treated as a precise prediction engine or as proof that any specific job will disappear by 2030. What Claude’s scenario modeling can do today Anthropic’s Claude budget futures use case provides the strongest first-party evidence of how scenario exploration works in practice. Claude can take budget context and produce three future scenarios side by side . The example includes stacked bar charts, a switch between dollar and percentage views, a short interpretation for each scenario, and follow-up prompts for refining constraints or extending the analysis. The page is presented as a finance use case, but its value is broader than finance teams. A manager can use the same planning pattern to test how assumptions affect a plan: for example, different levels of AI assistance in a r

Another week, another high-profile exit from a frontier AI lab. But this one is different in a way that matters for anyone who builds, deploys, or depends on large language models. Jacob Coxon, a 27-year-old pretraining researcher, announced late Tuesday that he has resigned from Anthropic and is leaving the AI industry entirely. He spent the last three years working on pretraining at both OpenAI and Anthropic, the two labs most often described as the current frontier. He did not leave quietly. His resignation thread on X named both companies as acting irresponsibly, and the Wall Street Journal ran an exclusive interview the same night. Here is what happened, and why it should register with technical people, not just policy watchers. What Coxon actually said The core of his statement is short. In his own words, both labs are "racing straight to self-improving superintelligence and gambling with our lives." He estimates a greater than 10 percent chance that advanced AI could kill all humans within the next decade, and he says that this belief is not fringe inside the labs. Colleagues, he claims, now talk openly about "crunchtime" and "endgame." Notably, he does not treat the two companies identically. Coxon's view, as reported by the WSJ and summarized across coverage, is roughly this: OpenAI, in his assessment, has not fully internalized the civilizational stakes of what it is building. Anthropic, he believes, does understand the stakes and is making earnest safety efforts, but is racing anyway on the logic that no competitor can be trusted to reach the frontier first. He moved from OpenAI to Anthropic earlier this year precisely because of Anthropic's safety reputation. His conclusion after working there is that the problem is structural, not cultural: no single company can responsibly build systems that outperform humans across a broad range of tasks without either government intervention or a coordinated slowdown across the industry. His timeline is aggressive. I

AI agents are moving beyond chat interfaces. In decentralized finance, they can monitor blockchain activity, analyze market data, identify opportunities, and potentially execute on-chain actions. But giving an AI system access to a blockchain wallet creates a serious security challenge. An agent that can only read blockchain data has limited risk. An agent that can sign transactions can potentially move funds, interact with smart contracts, or make irreversible decisions. That makes secure AI agent architecture especially important for DeFi applications. This guide explains how to design an AI-powered DeFi agent that can move from blockchain data to controlled on-chain actions while keeping security and human oversight at the center. What Is a DeFi AI Agent? A DeFi AI agent is a software system that combines artificial intelligence with blockchain infrastructure. Instead of simply responding to user prompts, an agent can observe information, reason about it, and perform predefined actions. A typical architecture may include: An LLM for reasoning Blockchain RPC providers Wallet or transaction infrastructure Smart contracts DeFi protocols Market and protocol data Risk-management rules Transaction simulation Monitoring and logging For example, a portfolio agent could monitor a user's positions and report: “Your lending position has crossed the configured risk threshold.” A more advanced system could prepare a transaction to rebalance the position. However, automatically sending that transaction should require additional security controls. Start With Read-Only Blockchain Access One of the safest ways to build a DeFi agent is to begin with read-only functionality. The agent can retrieve: Token balances Wallet positions Liquidity-pool data Lending positions Token prices Transaction history Smart-contract events At this stage, the AI does not need private keys or transaction-signing permissions. For example, a Python application can retrieve an ERC-20 balance through a blo

Instinct’s new email feature lets the AI agent create and manage accounts, contact businesses, handle support requests, and do more on users' behalf.

Anthropic researcher Jacob Coxon resigned over AI extinction fears, calling for pacing agreements between labs.

Users can ask the assistant to do things like "Create a cart for my Saturday tailgate for 25 people and include some brunch items," or "Build a cart for easy school lunches and after-school snacks," Shipt says.

A new class-action lawsuit accuses Anthropic of misleading advertising about Claude usage with its top-tier plan.

Nintendo just announced a new 2D entry in the Metroid series for the Switch 2, titled Metroid Ravenous. It's launching on January 28th, 2027. The new game looks like an evolution of Metroid Dread. While it's still a 2D game, it features 3D graphics, lots of cinematic action shots, parrying enemies, and Samus rolling around […]

Falling token costs, cheaper models, and less spend per employee—AI adoption isn't playing out the way hyperscalers hoped.

Nintendo just wrapped part two of its double feature this week. On Tuesday, the company hosted a showcase all about the Legend of Zelda franchise, including footage and a November 5th release date for the Switch 2 Ocarina of Time remake. On Wednesday, the company aired a new Nintendo Direct, which spotlighted Switch 2 games […]

Chris Lehane argues that stronger AI capabilities require stronger safety evidence, shared standards, and durable policy action while the policy window remains open.

Amazon's Prime Video is launching a new AI-powered feature that lines up an actor's mouth with "human-dubbed" audio. The feature is only available with the English dub of the German series Maxton Hall for now, but Prime Video plans to expand it to "additional titles" in the future. In its announcement, Prime Video says it […]

Instacart launches its own AI assistant that can make it easier to decide what to buy.

Kepler Computing claims a new approach to chip design—and a proprietary material—can help end the supply bottlenecks that have sent memory prices surging.

Cymphony was valued at more than $100 million in a $25 million Series A co-led by Sequoia and SMBC Fin Atlas Beyond Fund.