Ad
Skip to content
Read full article about: Salesforce releases new AI slackbot based on Anthropic's Claude

Salesforce has launched a new Slackbot built on Anthropic's Claude AI model. According to co-founder and Chief Technology Officer Parker Harris, the company is also testing alternatives. The AI assistant lives directly inside Slack and can search data across Slack, Salesforce, Google Drive, Box, and Atlassian's Confluence. It uses context from conversations, files, and channels to answer questions, create content, and prepare meetings, while respecting existing access rights and permissions.

Salesforce

Slackbot is now available for Business+ and Enterprise+ customers, with a gradual rollout running through February. Down the road, Slackbot will also be able to work with Agentforce and other AI agents.

Read full article about: OpenAI's "Sweetpea" AI wearable allegedly takes aim at Apple's Airpods

A new leak reveals details about OpenAI's planned hardware, an audio device designed to compete with Apple's Airpods. X and Weibo leaker "Smart Pikachu" claims OpenAI is developing a device codenamed "Sweetpea" with designer Jony Ive reportedly involved. The alleged September launch targets 40 to 50 million units in year one.

The device supposedly features an oval metal housing with two capsule-shaped components worn behind the ear, running on a 2nm chip with Samsung Exynos as the frontrunner. A separate chip would enable iPhone control through Siri. Material costs are reportedly close to smartphone level.

Die Komponenten von "Sweetpea": ein EMG-Signalfenster zur Erkennung von Muskelsignalen, ein Keramik-Hautkontaktfenster, Hauptplatine mit Lithium-Ionen-Akku sowie ein Ultraschall-Sende-/Empfangsmodul. | Bild: via zhihuipikachu
The leaked diagram shows "Sweetpea's" alleged components: EMG signal window, ceramic skin contact window, mainboard with lithium-ion battery, and ultrasonic transmitter/receiver module. | Image: via zhihuipikachu

If the leak proves accurate, Foxconn could produce up to five OpenAI devices by 2028, including a pen codenamed "Gumdrop." The manufacturer reportedly sees this as a chance to recover after losing all Airpods programs to Luxshare. OpenAI allegedly favored Luxshare initially but switched to Foxconn to enable production outside China.

Read full article about: Microsoft pledges to cover data center power costs as community pushback grows

Microsoft is rolling out a new initiative for AI data centers after facing mounting opposition from communities across the US. The company says it will fully cover the power costs of its data centers, ensuring residents won't see higher electricity bills as a result. The announcement comes as data center regions like Virginia, Illinois, and Ohio have seen electricity prices climb 12-16 percent faster than the national average.

Beyond power costs, Microsoft is making several other commitments: the company will stop requesting local tax breaks, cut water consumption by 40 percent by 2030, and replenish more water than it uses. Microsoft President Brad Smith told GeekWire that the industry used to operate differently and now needs to change its approach. Trump previewed the announcement on Truth Social before Microsoft made it official.

As part of the initiative, Microsoft also plans to train local workers and invest in AI education programs in affected communities.

Read full article about: AI models don't have a unified "self" - and that's not a bug

Expecting internal coherence from language models means asking the wrong question, according to an Anthropic researcher.

"Why does page five of a book say that the best food is pizza and page 17 says the best food is pasta? What does the book really think? And you're like: 'It's a book!'", explains Josh Batson, research scientist at Anthropic, in MIT Technology Review.

The analogy comes from experiments on how AI models process facts internally. Anthropic discovered that Claude uses different mechanisms to know that bananas are yellow versus confirming that the statement "Bananas are yellow" is true. These mechanisms aren't connected to each other. When a model gives contradictory answers, it's drawing on different parts of itself - without any central authority coordinating them. "It might be like, you're talking to Claude and then it wanders off," says Batson. "And now you're not talking to Claude but something else."

The takeaway: Assuming language models have mental coherence like humans might be a fundamental category error.

Comment Source: MIT
Read full article about: Anthropic's Claude Cowork was built in under two weeks using Claude Code to write the code

Anthropic's Claude Code inventor says his tool wrote almost all the code for Claude Cowork. Claude Cowork is a newly launched AI tool from Anthropic that builds on Claude Code but adds a user-friendly interface for non-programmers. According to Claude Code inventor Boris Cherny, "pretty much" all the code was generated using Claude Code.

Claude Code inventor Boris Cherny says his tool wrote nearly all the code for Claude Cowork. | Screenshot via X

Product Manager Felix Rieseberg says the app came together in just a sprint and a half, roughly one and a half weeks. The team had already built some prototypes and explored ideas beforehand, though, and the current release is still a research preview with a few rough edges, Rieseberg says. Claude Code also provided an extensive foundation to build on; Rieseberg is likely referring mainly to the front-end work.

UK startup turns planetary biodiversity into AI-generated drug candidates

UK company Basecamp Research has developed AI models together with researchers from Nvidia and Microsoft that generate potential new therapies against cancer and multidrug-resistant bacteria from a database of over one million species.

Read full article about: OpenAI acquires Torch to build a "medical memory for AI"

OpenAI is buying health app Torch for around 100 million dollars. The deal includes 60 million upfront and the rest in retention shares, The Information reports. Torch unifies scattered health records into what the founders call a "medical memory for AI", "a context engine that helps you see the full picture, connect the dots, and make sure nothing important gets lost." The app runs on OpenAI models. All four employees, including CEO Ilya Abyzov, are joining OpenAI.

The deal signals OpenAI's push toward a personalized health assistant in ChatGPT. Last week, the company launched a ChatGPT Health section and an offering for healthcare companies. Anthropic recently added health features to Claude as well. The moves reflect a shared bet on a massive market: hundreds of millions of weekly chatbot conversations already focus on health.

Read full article about: Apple turns to Google's Gemini as Siri's technical debt becomes too much to handle

Apple will use Google's Gemini models for its AI features, including a revamped version of Siri. The multi-year partnership means Apple will rely on Google's Gemini and cloud technology for its upcoming products, according to CNBC. The new features are expected to roll out later this year.

In a statement, Apple said that after careful evaluation, Google's technology offers the most capable foundation for its applications. Rumors about talks between the two tech giants first surfaced in March of last year. Later reports suggested the switch would cost Apple more than one billion dollars annually.

The move comes as Apple continues to struggle with Siri's underlying architecture. Internal reports describe Siri as a technically fragmented system built from old rule-based components and newer generative models - a combination that makes updates difficult and leads to frequent errors. Apple is also working on an entirely new in-house LLM architecture and a model with roughly one trillion parameters, aiming to eventually break free from external providers. Google faced similar challenges early on keeping pace with OpenAI's rapid progress but managed to catch up.

Comment Source: CNBC