AI/ML News & Innovations Hub

Comment on 87% ImageNet Accuracy, 3.8ms Latency: Google’s MobileNetV4 Redefines On-Device Mobile Vision by Hales

Sub-4ms latency with 87% top-1 is absolutely insane for an on-device model. Hardware-aware NAS is clearly paying off here. Optimizing directly for the EdgeTPU block layout makes a world of difference compared to just slapping a generic transformer stack on a mobile processor. I’ve been testing edge computer vision apps on budget unlocked phones from carriers like Gen Mobile and seeing this kind of performance trickling down to mid-range and entry-level mobile hardware is super exciting. Bringing high-accuracy, real-time vision processing on-device without torching the battery or relying on cloud inference latency is huge for actual real-world deployment.

[8월14일] AI는 싸졌는데 비용은 늘어난다…‘토큰 경제학’이 흔드는 AI 가격표

AI를 도입한 기업들이 갈수록 비용을 예측하기 어려워지고 있습니다. BBC에 따르면, 최근 AI 에이전트를 서비스에 적용하고 있는 기업들은 앞으로 1~3년의 AI 비용을 미리 계산하는 것조차 쉽지 않다고 밝혔습니다.우버가 올해 초 불과 몇 달 만에 1년 치 AI 코딩 토큰 예산을 모두 소진했다는 것은 잘 알려진 사실입니다. 마이크로소프트는 엔지니어들의 일부 외부 AI 코딩 도구 사용을 제한한 것으로 전해졌습니다.그러나 이는 단순히 AI가 비싸서 생긴 문제가 아닙니다. AI 자체의 가격은 계속 내려가고 있으며, 실제로 4개월여 만에

Don't classify. Hallucinate!

Don't classify. Hallucinate! I still have quite a bit of older content on my blog that I never got round to tagging. My blog has 1,856 tags - likely too many to feed to an LLM in one go and say "which of these tags match the following content". Doug Turnbull has a neat solution. Tell the model to output tags without any details of the existing vocabulary, then use vector embeddings against the existing corpus to find the concrete tags that are closest to the ones the model imagined might fit! His example prompt suggests including an example of the shape of your tags to help the model make a more useful guess: Your task is to create novel, never seen before, furniture, home goods, or hardware classification that best fit a search query. Product classifications might look like: Furniture / Living Room Furniture / Coffee Tables & End Tables / Coffee Tables Décor & Pillows / Decorative Pillows & Blankets / Throw Pillows Furniture / Bedroom Furniture / Dressers & Chests Kitchen & Tabletop / Kitchen Organization / Food Storage & Canisters School Furniture and Supplies / School Furniture / School Chairs & Seating / Stackable Chairs Baby & Kids / Toddler & Kids Bedroom Furniture / Kids Beds Here's the query to generate classifications for: brown coffee table Tags: search , ai , generative-ai , llms , embeddings , doug-turnbull

Feature Proposal: Emergency Assist – AI-Assisted Emergency Response

Hello, I would like to propose a feature that I believe could have a meaningful impact on user safety in the future. I envision a feature called Emergency Assist , designed to respond when a ChatGPT user may be in immediate danger or experiencing a life-threatening situation. This would not be limited to medical emergencies such as choking. It could potentially cover situations including: choking or sudden difficulty breathing, severe bleeding, symptoms of a stroke or heart attack, unconsciousness or sudden deterioration, car accidents, fires or gas leaks, poisoning, physical attacks or other immediate threats, situations where a person is unable to make an emergency call or speak. If ChatGPT detected from the conversation that the user might be in immediate danger, it could display a prominent “Call for Help” option. After confirmation from the user, the phone could initiate a call to the appropriate emergency number (such as 112 or 155, depending on the country). Where technically and legally possible, it could also provide emergency responders with basic information the user has shared during the conversation, such as the nature of the emergency and the user’s available location. I believe an especially valuable part of this feature would be a silent emergency assistance option . For example, someone being attacked might be hiding and unable to safely speak or make a phone call. In such a situation, ChatGPT could potentially allow the user to request emergency assistance…

Function calling + Assistants API: any pattern for capping cost when the model chooses to call a paid function repeatedly?

On your second question first: raw chat completions (or Responses) is easier, and it isn’t close. You’re the one running the tool loop there, so there is no interception step - you just don’t call the paid thing until you’ve checked the budget. Assistants puts its own run lifecycle in the middle. You can still catch it at requires_action before you submit_tool_outputs, but you’re working inside their orchestration rather than your own. For the confirmation part, don’t cancel the run when it goes over. Cancelling leaves the user with an answer that stops and nothing on screen saying why. We check the budget before executing, and if it’s over we submit a tool output like {“error”: “budget_exceeded”, “spent”: 4.10, “limit”: 5.00} and let the model explain it in its own words. The approval case is the same shape: hold the call, show the user the arguments and what it will cost, then either run it and submit the real result, or submit the refusal. Worth adding since the reply above is about migrating: all of this lives in your code rather than the API, so it carries over to Responses unchanged.

New York City Lawmakers Push to ‘Ban the Scan’ at MSG

At a press conference outside Madison Square Garden, politicians, musicians, and privacy advocates argued for tighter restrictions on how public venues deploy biometric surveillance.

Lawsuit alleges Sony investigated WPP's media operation and said it ran a 'global crime scheme'

A legal filing alleges Sony hired investigators to probe WPP, who said the agency improperly withheld rebates from clients.

Anthropic announces watermark detection API that will let third parties detect Claude's AI texts

Anthropic will soon offer a watermark detection API that lets third parties check whether text was written by Claude. The technology builds on Google's SynthID method and tweaks the randomness during word selection without affecting text quality, Anthropic says. The approach has limits with fact-heavy text, code, and heavy rewriting. The article Anthropic announces watermark detection API that will let third parties detect Claude's AI texts appeared first on The Decoder .

Feature request: Reference-based file management across Chats, Projects, and File Library

Problem: Users can’t easily tell which files are still “in use” across chats/projects, making cleanup risky and storage management inefficient. Proposal: Maintain explicit, global relationships (“references”) between files and every context that uses them (Chats, Projects, File Library). Reference tracking scope: Track all references to a file across the product; if no reference exists anywhere, the file becomes eligible for cleanup. What counts as a reference: Any file uploaded by the user or generated during chat conversations/tasks that is used/associated with a chat/project (and broader usage). Deletion behavior: When a chat/project is deleted, ChatGPT should detect files that now have zero remaining references and offer to delete them automatically; files still referenced elsewhere are preserved. Safety: Deleted files go to a temporary trash/bin for a retention period before permanent deletion. Retention period: Configurable user option (with a platform default). Include all reference sources: Yes—include references from Chats, Projects, File Library, Memory, custom GPTs, and other features. Benefit: Automatic cleanup based on actual usage (references) rather than file age; clearer separation between temporary working files and permanent reference material, especially alongside Projects, Memory, and the File Library. Thank you!

Anthropic defends and explains its AI watermarking feature

Anthropic addresses concerns about its new watermark, which will affect Claude users globally. It said the watermark is to comply with the EU AI Act.

GLM-5.3: How Chinese labs keep stride with the frontier

Hint: It’s really not a distillation story.

The Best Movies to Stream This Month (August 2026)

Avatar Aang, Heartstopper Forever, and Exit 8 are among the films deserving of your eyeballs this month.

Can OpenAI detect AI-generated images?

I am developing a website to eliminate the AI generate media in our website like other platforms using it so how to use API key of OpenAI in our website?

Feature Request: ChatGPT Usage Analytics - Statistics and Insights About Your Conversations

I think it would be really interesting to have a page where we could see some statistics about our own ChatGPT usage, kind of like Spotify Wrapped. For example, the total number of chats and messages, how many days we have used ChatGPT, our longest streak, average chats per day, most active days and hours, usage over time, most discussed topics, and maybe even which models we use the most. It could also have some simple insights based on our history, like: “You used ChatGPT on 80% of the days this year.” “Your most discussed topics are programming, technology and gaming.” “Your usage increased by 30% compared to last year.” I think this could also work really well as an annual “ChatGPT Wrapped”, showing a summary of how we used ChatGPT throughout the year. The data is already part of our ChatGPT history, so it would be really interesting to have a way to actually see and analyze it. I would definitely use this. If you need some help building it, you can call me

The Next Big Influencer Is This 4-Foot-Tall Robot From China

The Unitree G1 has found online fame as a relatively affordable robot that can charm a crowd. But can it ever hold down a real job?

Provider-independent identity and delegated authority for external AI agents

I’ve been working on a provider-independent governance boundary for external AI agents and would value technical criticism from people building with OpenAI’s agent stack. The premise is simple: Authenticating an artificial agent should establish who is acting—not automatically establish what it is authorized to do. The prototype separates: external-agent identity from authority narrowly scoped and time-bounded delegation deterministic execution/approval boundaries governed persistent state with provenance action receipts linking principal → delegation → agent → workflow → action → result revocation and recovery while preserving the evidence chain The architecture is: External Agent Provider → Governed Agent Adapter → Deterministic Governance Kernel In the current executable demonstration, a simulated external agent begins authenticated but with zero ambient authority. It then receives narrowly bounded capabilities, encounters execution/approval/denial boundaries, produces deterministic evidence, and is finally revoked while the prior evidence chain remains intact. Important limitation: the provider transport is currently deterministic and simulated. I am not claiming a live OpenAI, Codex, Grok, or other provider integration. The goal is not to replace provider-native authorization. It is to investigate what governance should exist around increasingly capable provider-supplied workers when identity, delegated authority, persistent state, execution evidence, and recovery need…

Zed Launches Delta, a Standalone App Built for AI Agents Writing Your Code

Zed v1.15 ships smarter Git diffs and JSX improvements while Delta, a new multiplayer agent coding environment, enters private beta

Nous Research's Hermes Agent Ships a Loop Command That Watches CI for You

Hermes Agent's new /loop command lets you re-run any prompt on a recurring schedule, with smart self-pacing built in

Google releases C++ library for content provenance and authenticity

Google has introduced Credentio, an open source C++ library for C2PA (Coalition for Content Provenance and Authority) Content Credentials. Announced August 13 and available at the mediaprovenance repository, Credentio provides an API designed to run locally within developer applications. This removes the need to send media files to cloud servers for validation, which incurs privacy, latency, bandwidth, and file size limitations. Credentio is designed to start working with C2PA specification versions 2.2 and 2.4. This is the same code that has powered nearly 40 different conformant C2PA-enabled Google products to scale to tens of billions of generated assets, including images, videos, audio files, and documents across many file formats, Google said. Through Credentio, media files do not need to be transmitted back to Google or external validation endpoints while offering zero bandwidth overhead, instant validation verdicts, and complete data privacy, the company said. By validating media files locally Credentio eliminates external data transmission requirements, minimizes verification latency, and delivers immediate results even in high-throughput workflows. And media contents remain securely within the local environment. Credentio is tailored specifically for developers seeking to build performant, enterprise-grade C2PA validator products. Its small memory footprint makes it ideal for integrating into resource-constrained client applications, server pipelines, or high-perfor…

Why does everything feel meh? We're in the era of joyless decadence – Stateside with Kai and Carter

We should, in theory, be happy. Thanks to our phones, we can do practically anything – listen to our favorite bands' entire discographies, have extravagant meals delivered to our door, travel the world through an endless scroll. So why do so many of these experiences feel so joyless? Why do so many of us feel so empty all the time? Hosts Kai Wright and Carter Sherman wrestle with these questions with academics Michèle Mendelssohn and Charlie Tyson , who argue that we are living in an era of decadence without pleasure. While decadence always arises in times of extreme inequality, they say that tools like AI and GLP-1s reveal how we've lost our ability to truly enjoy things – and, ironically, reclaiming decadence may help us finally have fun. Michèle Mendelssohn and Charlie Tyson’s Guardian essay: Why does everything feel joyless? Welcome to the age of decadence without pleasure Continue reading...

Self-driving trucks are officially testing on California highways

Aurora Innovation and Kodiak AI, two companies developing self-driving trucks, have received permits from the California Department of Motor Vehicles.

White House prepares to host crypto and prediction market execs

Attendees will include executives from Coinbase, a16z, Ripple, and Chainlink, plus Kalshi and Paradigm, as well as trade association Digital Chamber, according to people familiar with the plans.

DeepSeek's innovative harness treats everything as a plug-in

Chinese AI labs keep moving forward while US labs play defense

Chinese drone maker DJI gets another shot at Pentagon blacklist removal

A US appeal court on Friday gave Chinese drone maker DJI another chance to challenge its inclusion on a Pentagon blacklist of companies allegedly linked to the Chinese military. DJI, a Shenzhen-based drone maker, has been engaged in a years-long legal battle with the US Department of Defence to be removed from the list. The latest ruling from the US Court of Appeals for the District of Columbia Circuit did not remove DJI from the list, but sent the case back to the lower court to reconsider one...

Migrant charity SOS Méditerranée says 11 bodies spotted in sea off Libya's coast

The EU border agency Frontex said the total number of irregular border crossings into the European Union fell by 26% in 2025 from the previous year to almost 178,000.

Pika Labs Launches Pika Audio at 9x Cheaper Than ElevenLabs

Pika launches four generative audio models — SFX, Speech, Music, and Soundtrack — priced up to 20x cheaper than ElevenLabs and other rivals

EU pushes back as US revives trade pressure on bloc and accuses Brussels of enabling China to evade tariffs

The EU already dialled back the scope of the environmental and human rights supply chain standards laws under pressure from businesses, but the US said that did not go far enough.

I made chef Alex Guarnaschelli's favorite 'accidentally vegan' dish. The recipe is now my go-to for an easy dinner.

I don't follow a plant-based diet, but I loved this celebrity chef's affordable and easy-to-make vegan dinner recipe for cauliflower "steaks."

Jason Arday, former Cambridge University professor at centre of plagiarism row, found dead at home

Arday made headlines when the university appointed him as its youngest black professor in 2023 but he since faced snowballing claims he plagiarised parts of his PhD thesis.

EV prices are rising again as automakers pull back on discounts

Car companies slashed thousands off their EV prices as consumer interest waned. Now, prices are slowly inching back up.

Thrive’s Joshua Kushner chides Silicon Valley VCs over AI euphoria

The AI opportunity is huge, but "it would also be a grave error in our minds to let excitement weaken our investment discipline," Kushner warns in his first-ever investment letter.

X reveals the formula for going viral on its platform

X released the source code for its algorithm on Thursday. The details reveal how posts end up on the For You page.

Comment on Baidu Announces Autonomous Taxi & Bus Fleet for Hunan by Mills

Baidu’s plan is a fascinating step toward making autonomous transport part of everyday city life. The combination of L4 vehicles, geofenced routes, and V2X infrastructure shows that successful self-driving systems require more than smart cars; they also need reliable roads, communication, and safety systems. I’m especially interested in how autonomous buses could improve predictable routes and public mobility. It will be interesting to see how passenger confidence develops as these services become more common. For comparison, dependable route planning is also important in traditional passenger transport, such as https://pakumrahbusservice.com/, where reliability remains a key part of the travel experience.

Sourav Ganguly breaks silence over alleged death threats

Samsung has new Galaxy headphones in the works

Strings of code in Samsung's Galaxy Wearable app hint at an upcoming pair of over-ear headphones that could compete with the AirPods Max, SamMobile reports. Samsung's reportedly referring to the headphones as the "Galaxy H1," and SamMobile says they could launch sometime in 2027. That would make these the company's first pair of over-ear headphones […]

'I shot Mr Thompson': Luigi Mangione admits killing UnitedHealthcare CEO in shock court confession

'I shot Mr Thompson': Stunning confession in US health exec's slaying

What we know about the alleged Iranian hacks on US water utilities

Over the last couple of weeks, hackers have targeted and broken into the systems of several water plants in the United States. Here’s what we know and don’t know about this wave of attacks allegedly carried out by the Iranian government.

NVIDIA's NeMo Switchyard Cuts Agent AI Costs by 74% With Smart Model Routing

NVIDIA's open-source NeMo Switchyard routes each agent workflow step to the right model automatically, cutting costs by up to 74% without rebuilding your app

Read-it-later app Pocket shut down — here are the best alternatives

Pocket users have until October 8, 2025, to export their saved articles and other items, including lists, archives, favorites, notes, and highlights.

Russia admits new petrol shortages amid ongoing long-range Ukrainian strikes

Ukraine has been targeting Russia's energy sites with drones and missiles for months, hoping to hit a vital source of revenue for the Russian army.

Claude users are canceling their subscriptions, citing Anthropic’s new AI watermark

One developer said they feared Anthropic's watermark could lead clients to flag their work as AI-generated and raise questions about authorship.

A RAMageddon guide to back-to-school laptop shopping

If you’re a student looking for an affordable laptop, I have bad news and I have good news. The bad news is that computer prices are out of whack due to the ongoing RAM and storage crunch, making some a little pricier than they used to be and others wildly more expensive. The good news? […]

ANPD suspende lives no Discord

No dia 12 de agosto de 2026, a Agência Nacional de Proteção de Dados emitiu medida preventiva que suspendeu o recurso de transmissão ao vivo da plataforma Discord. A decisão é a primeira em que a Agência exercita sua competência fiscalizatória a partir do ECA Digital. Este texto tem o objetivo de destrinchar e analisar tecnicamente os elementos que levaram à decisão da ANPD. O post ANPD suspende lives no Discord apareceu primeiro em Data Privacy Brasil Research .

VCs are investing more money in social media startups this year after a dismal 2025

The number of venture capital deals backing social media startup has declined since 2021. The sector is seeing new signs of life, though.

British cyclist Finlay Tarling, 19, dies in Tour of Portugal crash

British cyclist Finlay Tarling, 19, dies in Tour of Portugal crash: organisers

heise+ | Wie der DWD mit KI, Satelliten und Flugzeugen bessere Wetterprognosen erstellt

Seit 65 Jahren schon macht der Deutsche Wetterdienst Wettervorhersagen. Vieles lässt sich mittlerweile gut berechnen, doch ein Phänomen sorgt für Probleme.

New Claims In WPP Whistleblower Suit Allege Sony Probe Found Rebate Fraud

The ex-WPP executive suing his former employer for retaliation in a high-profile case alleging he was fired after trying to expose an illegal media rebate scheme now alleges that WPP […]

Tech analyst Ben Thompson dismisses the 'clearly absurd' concept embedded in AI watermarking

Thompson took particular issue with Anthropic adding a watermark even if Claude was used only for proofreading.

I visited the Obama Presidential Center, the most expensive presidential library in US history. Here are 15 things that surprised me most.

The Obama Presidential Center is the most expensive presidential library in US history. Its 19-acre campus houses a museum and other community spaces.

Digital Signal Processing Pioneer Bede Liu Dies At 91

Bede Liu, a digital signal processing pioneer, died on 7 May. He was 91. Liu was widely regarded as one of the founders of modern digital signal processing, a field that applies mathematical algorithms to analyze, modify, and transmit signals including sound, images, and video. The IEEE Life Fellow taught electrical engineering at Princeton for more than 50 years. From 1994 to 1997, he chaired the university’s electrical and computer engineering department . Liu’s research aided the transition from analog to digital processing of sound, images, and video. His work helped establish many of the mathematical and engineering techniques that underpin modern communications, multimedia systems, and consumer electronics. Although little known outside engineering circles, his work is embedded in technologies used by billions of people. The low-power digital signal processors that make cellphone calls, streaming video, and Internet communications possible can be traced to research he conducted in the 1970s and ‘80s. Liu received the 2018 IEEE Jack S. Kilby Signal Processing Medal for “sustained contributions to the analysis and the development of low-complexity realizations of digital signal processing algorithms.” “We stream music and video. We take photos with our phones, and we send them around. We don’t even think about it,” IEEE Life Fellow H. Vincent Poor said in an obituary for Liu . “But it’s all because of the signal processing, image processing, and video processing that’s b…