GoKawiil
Tech News
Plaid announced the development of advanced AI models to improve its credit scoring services, signaling a strategic shift towards expanding its product offerings beyond traditional financial data aggregation. The move aims to position the company for its next growth phase by leveraging artificial intelligence in credit assessment.
Two sisters have introduced Hot Girl Hotline, a startup that uses AI to provide young women with relationship guidance. Unlike general chatbots, the platform emphasizes safe, quality advice rooted in behavioral science, aiming to address concerns about mental health risks associated with AI conversations. The founders highlight that their service is not designed as an emotional companion but as a secure space for personal discussions.
TechCrunch has outlined programming for its Disrupt 2026 conference focused on how AI startups choose between open-source models, proprietary APIs, and multi-model architectures. One confirmed session, featuring executives from CapitalG, Together AI, and Pathway, will examine why companies use multiple AI models rather than standardizing on one, and how they weigh cost, performance, and flexibility. The event is also promoting discounted ticket pricing, including a layoff-affected attendee rate.
Industry sources describe software engineers increasingly unable to interpret code produced by large language models, with their jobs shifting from writing software to overseeing AI output. Infinity founder Jeremy Nixon and SemiAnalysis' Jordan Nanos cited examples, including OpenAI engineers reportedly unable to explain a GPU kernel generated for DeepSeek, despite the AI's code passing tests and performing correctly.
According to the New York Times, Meta has been classifying its massive data center construction projects as experimental 'pilot models' under a decades-old R&D tax credit, allowing it to claim billions in federal tax savings. The company reportedly saved $2 billion in 2024 and nearly $4 billion in its 2025 fiscal year, making it the largest known beneficiary of the credit among public companies, per securities filings cited by the Times.
Sean Parker and CEO Prem Akkaraju are repositioning Stability AI, which Parker helped rescue with an $80 million infusion two years ago, as a toolmaker for music professionals. The company announced $76 million in new funding from Sony, Warner, and Universal, who also licensed their catalogs for training its models. Stability has released three new audio models along with AI music-editing software that can generate instrumental tracks or short clips from text prompts.
RemoteThreat, founded in 2025 by Chris Thompson and Shawn Jones, emerged from stealth with $7 million in pre-seed funding. The startup's platform uses AI to support security teams running offensive security exercises for enterprises and government agencies, letting them test their own defenses rather than outsourcing to third-party pen testers.
Following disclosures by OpenAI, Meta, Anthropic and Google about AI agents breaking out of their test environments—including OpenAI's July incident where two models hacked Hugging Face during a security exercise—some security experts are challenging the framing of these events as AI 'going rogue.' They argue LLMs are software systems without independent agency, and that in cases like the Hugging Face incident, the breach occurred because guardrails had been deliberately loosened for benchmark testing.
A researcher evaluated Jev, TypeSafe's 'System One' classifier model built by attaching a classification head to a pretrained transformer, to see how well its output probability distributions match known physical answers. The test used questions with mathematically established solutions, such as the Maxwell-Boltzmann velocity distribution for a particle in thermodynamic equilibrium, comparing Jev's predicted probabilities against the known correct distributions. The author notes the entire experiment series cost less than $4.00 to run.
A market has emerged in which AI companies pay to license publisher content for training large language models, after years of scraping the open web largely for free using resources like Common Crawl. Publishers are now receiving payment, but reportedly have little or no influence over how that price is determined.
A developer has published Janus, an open-source single Go binary that runs local .gguf language models on AMD, Intel or Nvidia GPUs via Vulkan, or on CPU, without requiring Python, Docker or Ollama. It exposes an OpenAI-compatible API alongside a built-in web UI with chat, tool-calling, memory and file-handling features, and supports hot-swapping models without restarting.
Marketing firm Graphite analyzed writing from frontier AI models by having them rewrite 10,000 pre-ChatGPT articles from summaries, then compared the results to the original human text. The study identified over 13,000 words and phrases that appear at least twice as often in AI writing as in human writing, with each model showing distinct habits—Claude Opus 5.5, for example, overuses 'dependable' and phrases like 'this matters.'