0:08
Small AI Models Gain Traction Around the World
What surprised me most was how the whole system got pulled off the cloud and onto a phone in a couple of hours. Alonge’s RxScanner was choking on a five‑minute round‑trip to a U.S. server, so his team pruned the model until it could run locally on an Android device, turning a bandwidth nightmare into an instant readout. That trick isn’t just a hack for pills; it’s the blueprint for “small AI” everywhere—tiny language models, often under a few billion parameters, that can live on phones, Arduinos or solar‑powered drones. By trimming or distilling big models, developers keep the core skill set while shedding the bulk, letting devices diagnose counterfeit drugs, spot diseased crops, or flag mosquito breeding sites without ever touching a data center. As hardware gets more efficient and chips with neural‑processing units become common, these edge‑first models are spreading fast, promising practical AI for the parts of the world where big‑scale cloud power simply isn’t an option.
0:59
Your family’s $300 stake in OpenAI
So, the weird part under the hood is that Sam Altman is actually sitting down with the Trump team to talk about handing the U.S. a 5 percent slice of OpenAI. It’s not a brand‑new idea—he’s been nudging at a “public‑share” model for years—but the concrete number and the presidential link are fresh.
If the deal went through, that 5 percent would be worth roughly $42 billion at today’s valuation. Split across about 133 million households, each would end up with a $320 equity stake. The plan isn’t to hand out the stock directly; instead the government would likely park it in a fund and let the returns trickle down, which could boost the payout if the company ever turns a profit.
The motive, beyond the optics, is twofold. First, it gives a kind of back‑pay to the creators whose work fuels the models. Second, it offers a safety‑net that might calm the anxiety about AI displacing jobs, even if economists aren’t convinced a profit surge is guaranteed.
All of this is still more story than policy. Altman’s pitch has been floating around for half a decade, and even the more aggressive Sanders proposal looks unlikely to materialize. What it does show is how unsettled the rules are around AI wealth, and how much the industry is banking on the idea that the boom will be big enough to share.
2:05
Microsoft lays off nearly 5,000 employees across Xbox, commercial sales
Microsoft cut around 4,800 roles, or 2.1% of its global workforce, on Monday — the latest in a series of layoffs that’s stoking fears of AI replacing jobs. The layoffs will hit Xbox and commercial sales the hardest.
2:21
If you use Google, you’re training its AI. Here’s how to opt out.
Consider this a belated PSA: A recent change to Google’s privacy settings is allowing the company to store more of your data, including media such as “images, files, and audio and video recordings,” to improve its AI models.
2:37
The ‘first’ AI-run ransomware attack still needed a human
An AI agent carried out the technical execution of a real-world ransomware attack for the first known time, but new details show a human still chose the victim, set up the infrastructure, and supplied stolen credentials — meaning it wasn't quite the fully autonomous cybercrime debut that last week's headlines suggested.
2:57
GPT-4's dominance lasted a year while today's top models barely survive seven weeks at the top
OpenAI's GPT-4 led the Epoch Capabilities Index for about a year, far longer than any model since. Since Claude 3 Opus took the top spot in February 2024, the lead has changed hands 17 times, with a median stay of just seven weeks. Competition is fiercer now, but the capability gains between models are shrinking.
The article GPT-4's dominance lasted a year while today's top models barely survive seven weeks at the top appeared first on The Decoder.
3:25
Tencent releases Hy3 open-source model that allegedly matches models up to five times its active size
Tencent has released Hy3, an open-source language model with 295 billion parameters built on a mixture-of-experts architecture. Only 21 billion parameters are active at any given time. Tencent says Hy3 matches models two to five times its size while cutting its hallucination rate in half to 5.4 percent.
The article Tencent releases Hy3 open-source model that allegedly matches models up to five times its active size appeared first on The Decoder.
3:53
Zhipu AI launches ZCode to challenge Claude Code and OpenAI Codex at a fraction of the cost
Zhipu AI is bringing GLM-5.2 to its ZCode development environment, pitching the model's long-context capabilities for complex coding tasks. New customers get a free five-day trial with up to 5 million tokens per day, and subscribers receive about 1.5 times more token quota through July 2026.
The article Zhipu AI launches ZCode to challenge Claude Code and OpenAI Codex at a fraction of the cost appeared first on The Decoder.
4:20
Cloudflare replaces its blanket AI bot block with granular controls for search, training, and agent crawlers
Cloudflare is giving all customers granular AI bot controls. Site owners can now manage Search, Training, and Agent bots separately instead of blocking them all at once. Starting September 15, 2026, Training and Agent bots will be blocked by default on ad-supported pages.
The article Cloudflare replaces its blanket AI bot block with granular controls for search, training, and agent crawlers appeared first on The Decoder.
4:48
Validating the RAG Answer Before the User Sees It: Spans, Quotes, and the Feedback Loop
Enterprise Document Intelligence [Vol.1 #8C] - Structured output is the start of validation, not the end: check the evidence, accept not-found, loop the feedback
The post Validating the RAG Answer Before the User Sees It: Spans, Quotes, and the Feedback Loop appeared first on Towards Data Science.
