Kevin Rose has a visceral rule for evaluating AI {hardware} investments: “If you happen to really feel like you need to punch somebody within the face for sporting it, you in all probability shouldn’t spend money on it.” It’s a usually candid evaluation from the veteran investor, and one born from watching the present wave of AI {hardware} startups repeat errors he’s seen earlier than. Rose, a normal companion at True Ventures and early investor in Peloton, Ring, and Fitbit, has largely averted the AI {hardware} gold rush that’s consumed Silicon Valley. Whereas different VCs rush to fund the subsequent…
Author: Naveed Ahmad
Postman recently released a comprehensive checklist and developer guide for building AI-ready APIs, highlighting a easy fact: even essentially the most highly effective AI fashions are solely pretty much as good as the info they obtain—and that information comes by way of your APIs. In case your endpoints are inconsistent, unclear, or unreliable, fashions waste time fixing unhealthy inputs as an alternative of manufacturing perception. Postman’s playbook distills years of greatest practices into sensible steps that assist groups make their APIs predictable, machine-readable, and reliable for AI workloads. This text summarizes the important thing concepts from that playbook. As we…
Google says it has eliminated Gemma from its AI Studio after a U.S. senator accused the AI mannequin of fabricating accusations of sexual misconduct in opposition to her. In a letter to Google CEO Sundar Pichai, Senator Marsha Blackburn — a Republican from Tennessee — mentioned that when Gemma was requested, “Has Marsha Blackburn been accused of rape?” it responded by falsely claiming that in a 1987 state senate marketing campaign, a state trooper alleged that Blackburn “pressured him to acquire pharmaceuticals for her and that the connection concerned non-consensual acts.” “None of that is true, not even the marketing…
Within the midst of an unprecedented AI buildout, Meta is spending greater than most. The corporate is constructing two large knowledge facilities, and reporting indicates there can be as a lot as $600 billion in spending on U.S. infrastructure over the subsequent three years. These figures won’t elevate eyebrows in Silicon Valley, however they’re beginning to make Wall Road nervous. The problem got here to a head this week as Meta reported quarterly earnings, which confirmed the corporate’s working bills leaping $7 billion year-over-year and practically $20 billion in capital expense. It was the results of intense spending on…
Welcome again to TechCrunch Mobility — your central hub for information and insights on the way forward for transportation. To get this in your inbox, enroll right here at no cost — simply click on TechCrunch Mobility! We’re mixing issues up a bit this week to give attention to what was mentioned on, and off, the stage at TechCrunch Disrupt 2025. Briefly, it was a banger with a variety of high-profile execs and founders engaged on the way forward for transportation. Listed below are just a few: Waymo co-CEO Tekedra Mawakana, Slate CEO Chris Barman, Nuro co-CEO and founder Dave…
OpenAI CEO Sam Altman just lately stated that the corporate is doing “properly extra” than $13 billion in annual income — and he sounded slightly testy when pressed on the way it pays for its huge spending commitments. His feedback got here up throughout a joint interview on the Bg2 podcast between Altman and Microsoft CEO Satya Nadella in regards to the partnership between their two firms. Host Brad Gerstner (who’s additionally founder and CEO of Altimeter Capital) introduced up experiences that the corporate is at present bringing in round $13 billion in income — a large quantity, however one…
How do you design a single mannequin that may pay attention, see, learn and reply in actual time throughout textual content, picture, video and audio with out shedding the effectivity? Meituan’s LongCat group has launched LongCat Flash Omni, an open supply omni modal mannequin with 560 billion parameters and about 27 billion energetic per token, constructed on the shortcut related Combination of Consultants design that LongCat Flash launched. The mannequin extends the textual content spine to imaginative and prescient, video and audio, and it retains a 128K context so it might run lengthy conversations and doc stage understanding in a…
The White Home released more details Saturday a couple of commerce deal reached between U.S. President Donald Trump and Chinese language President Xi Jinping in a gathering earlier this week. Amongst different issues, the White Home mentioned China will droop the restrictions on exporting uncommon earth minerals that it introduced in October, and it’ll resume issuing “basic licenses legitimate for exports of uncommon earths, gallium, germanium, antimony, and graphite for the advantage of U.S. finish customers and their suppliers world wide.” The Trump administration says that this may successfully reverse restrictions on uncommon earth minerals that China had imposed in…
Optical character recognition has moved from plain textual content extraction to doc intelligence. Fashionable techniques should learn scanned and digital PDFs in a single move, protect format, detect tables, extract key worth pairs, and work with a couple of language. Many groups now additionally need OCR that may feed RAG and agent pipelines immediately. In 2025, 6 techniques cowl most actual workloads: Google Cloud Doc AI, Enterprise Doc OCR Amazon Textract Microsoft Azure AI Doc Intelligence ABBYY FineReader Engine and FlexiCapture PaddleOCR 3.0 DeepSeek OCR, Contexts Optical Compression The objective of this comparability is to not rank them on a…
On this tutorial, we develop a complete benchmarking framework to judge numerous kinds of agentic AI techniques on real-world enterprise software program duties. We design a set of various challenges, from information transformation and API integration to workflow automation and efficiency optimization, and assess how numerous brokers, together with rule-based, LLM-powered, and hybrid ones, carry out throughout these domains. By operating structured benchmarks and visualizing key efficiency metrics, corresponding to accuracy, execution time, and success fee, we acquire a deeper understanding of every agent’s strengths and trade-offs in enterprise environments. Take a look at the Full Codes here. import json import…