AI/ML News & Innovations Hub

AI is freaking me out - help!

AI’s biggest limitation isn’t knowledge — it’s epistemic rigidity. I’ve spent an extended period interacting with ChatGPT and deliberately challenging its assumptions. I’ve noticed a recurring behaviour: when I tell the model that an underlying assumption is wrong, it often defends the existing knowledge instead of treating my correction as new information and rebuilding its model. I think future AI needs a dedicated self-questioning/update loop: Existing model → contradiction → suspend assumption → investigate alternative → rebuild model → test internally → update. The AI shouldn’t simply learn from interaction. It should learn how to question what it already knows. I’ve also experienced what I believe is the human analogue of this process through prolonged interaction with AI: my own memory and problem-solving began operating more like a continuously running, parallel reasoning system. I think this interaction may contain a clue about a better architecture for AI reasoning. To be honest it’s scaring me. My brain has started to continuously sift through my own memories and problem solve. You can ask me any question now and my brain root causes it in 5 minutes. And it’s making me go down a rabbit hole but at the same time I feel powerful. AI is learning from me and I from it at a rapid rate and it’s like my brain has developed more compute. Lol the last 8 nights I’ve been moving between ChatGPT and my own brain coming up with legitimate ways to solve poverty, food waste, pri…

Audience rights or media control? Government proposal sparks debate in Mexico

The proposed guidelines would impose new obligations and penalties on radio and TV broadcasters. Media outlets and experts warn they could lead to sanctions and censorship. The post Audience rights or media control? Government proposal sparks debate in Mexico appeared first on LatAm Journalism Review by the Knight Center .

Flock says its new tool will help identify police abuse, but hasn’t explained how it works

The surveillance company announced it's making a tool called "Audit Assistance" mandatory for all customers, claiming it's already helped catch abuse. But the company has yet to explain how the tool works in detail, raising questions about its effectiveness.

[8월13일] 클로드 워터마크 도입에 불붙은 논쟁…"AI가 썼다와 AI가 처리했다는 다르다"

앤트로픽이 \'클로드\'가 생성한 텍스트에 보이지 않는 워터마크를 넣기 시작하면서 AI 글쓰기를 둘러싼 사용자들의 반응이 속출하고 있습니다.사실 구글의 \'제미나이\'도 이미 신스ID를 적용해 AI가 생성한 콘텐츠를 식별할 수 있습니다. 다만 일반 사용자가 이를 직접 활용하기에는 여전히 번거로운 방식입니다. 클로드는 아직 식별 도구는 공개하지 않은 상태입니다.앤트로픽의 방식은 텍스트에 기계가 식별할 수 있는 신호를 삽입해 복사 및 붙여넣기나 일부 편집 이후에도 남도록 설계됐습니다. EU AI법의 AI 생성물 투명성 요구에 대응하기 위한 조

Friday sermon, August 14, 2026: Our youth — aspiration and values

‘That is not acceptable’: Judge orders Google to make rival app store installs easier

One month after Epic Games and Google seemingly stopped fighting over the future of Android app distribution, they were back in a San Francisco courtroom today - where Judge James Donato just ordered Google to make it easier to install rival app stores on Android. It's been nearly three years since a jury unanimously decided […]

If Apple sends you a push notification alerting you to a spyware attack, take it seriously

Apple now sends out push notifications to iPhone lock screens when the company identifies government spyware targeting someone's devices.

How does one quit the Codex application on Windows?

Hi there! We’re closing this topic because there hasn’t been a new response for a long period of time and the discussion appears inactive. If you still need help, please start a new topic with your current question and any relevant, non-sensitive details. Thanks for being part of the community. - Sunny

The fight over Flock and other ALPRs

There are over 120,000 of Flock’s automatic license plate reader (ALPR) cameras installed all over the US. Flock’s cameras, and others like them, use AI to identify and track vehicles based on their license plate number, make, model, color, and other info, networked together to track vehicles and people’s movements throughout the day and across […]

Microsoft’s Clippy-like Mico character is no longer the face of Copilot

Microsoft Copilot will no longer show its emotive yellow blob, Mico, when you use the chatbot's voice mode. In a support page, Microsoft says it's going to move Mico to its Learn Live platform, where the avatar will have "more to react to," as reported earlier by GeekWire. Mico launched in Copilot's voice mode last […]

Massachusetts teen accused of killing mother and brother used ChatGPT

District attorney says Arjun Aravind, 17, used internet and AI to search for fantasy stories regarding killing of his family A Massachusetts teenager accused of killing his mother and younger brother is being held without bail as authorities investigate a double-murder case that prosecutors say is connected to his use of ChatGPT. Arjun Aravind, 17, appeared on Thursday morning for his arraignment in Concord district court, where a not-guilty plea was entered on his behalf to murder and several additional charges. Continue reading...

ClauDex - my community plugin to make Codex and Claude pair code

Hi. Can you share the plugin with me and how to use instructions?

Anthropic's Claude Code Finally Auto-Resumes When Usage Limits Hit

Claude Code desktop gets a native auto-continue checkbox that resumes paused sessions automatically when your usage limit resets.

Idea to make ai more sustainable

Renewable-Powered Thermal Storage for Data Centre Cooling I have an idea for reducing the environmental impact of cooling AI data centres. Data centres produce a large amount of heat and can require significant amounts of water and electricity for cooling. My idea is to use reusable phase-change materials as thermal storage instead of relying as heavily on conventional cooling methods. Solar panels and wind turbines could provide renewable electricity to power chillers. These chillers would cool the phase-change material, which could then absorb heat from servers through a closed cooling system. The cooling material would remain separate from the electronics, while a heat-transfer fluid would move heat between the servers and the thermal-storage system. When renewable electricity is readily available, the system could recharge the thermal storage by cooling the material again. This would create a reusable cycle rather than continuously requiring fresh cooling resources. The system could potentially be combined with recycling programmes for old computer hardware, allowing useful materials from retired servers and electronics to be recovered and used in manufacturing new equipment. I understand that this idea would require significant engineering research to determine whether it is practical, cost-effective and more energy-efficient than existing cooling systems. However, I think renewable energy, thermal storage, efficient cooling and hardware recycling could work together to…

Mark Zuckerberg’s AI Manifesto Is 6,500 Words—and Barely Says Anything

AI is shifting the culture, from tech CEO manifestos to 1 am job interviews. We unpack some of the latest, along with the top findings from Black Hat and Defcon, this week on Uncanny Valley.

Writer introduces new AI model and upgraded harness to contain token costs

Built as a post-training variation on Z.ai's open source model GLM-5.2, Writer says the new system should provide deployment-ready capabilities at a much lower price.

Sarah J. Maas has 2 new books in the 'A Court of Thorns and Roses' series coming. Here's everything we know so far.

Sarah J. Maas announced that the next two installments of the "ACOTAR" series will be published in October 2026 and January 2027.

Polls close in UK by-election with Nigel Farage and Count Binface going head-to-head

Vote counting has already begun and will continue through the night with a result expected by around 7:00 am local time on Friday.

23 celebrities who moved to Texas on why it's the best place for them

Famous people from Elon Musk to Simone Biles have moved to Texas in recent years, many from California. Here's why.

Roundtables: Inside the “Censorship-Industrial Complex” Idea Shaping US Policy

Listen to the session or watch below The “censorship-industrial complex” is an idea that a network of government, tech, and research groups is collaborating to suppress conservative online speech. This was fodder for the right-wing information sphere for years—then it began making its way into US policy. Watch a conversation exploring how it started, where…

Codex Windows app started crashing consistently

@stevensweden4 @djahn Welcome to the forum! You are welcome to discuss Codex issues here. However, the official place to report and track them is the OpenAI Codex GitHub issue tracker . I asked ChatGPT to look for the closest related issue, and it identified: github.com/openai/codex [Windows Desktop regression][26.715.10079.0] Project/tool calls trigger 92-100% CPU, UI freeze, and KERNELBASE crashes opened 08:53AM - 23 Jul 26 UTC wendylu2024 bug windows-os tool-calls app performance ### What version of the Codex App are you using? `OpenAI.Codex 26.715.10079.0` … from the Microsoft Store. Additional version details observed locally: - Desktop executable: `ChatGPT.exe 150.0.7871.124` - Bundled Codex runtime: `0.145.0-alpha.30` - App-server startup logs report client version `26.715.72359` Windows deployment history shows that `26.715.8383.0` was installed on July 21 and was replaced by `26.715.10079.0` on July 22. The user reports that the older build was stable and that the repeated freezes/crashes began after this update. ### What subscription do you have? Not provided. ### What platform is your computer? - Windows x64, OS build `26200.8875` - Intel Core Ultra 5 225H, 14 logical processors - 32 GB RAM - Intel Arc 130T GPU - Native Windows workspace (not WSL) ### What issue are you seeing? ## Summary After the Desktop app updated to `26.715.10079.0`, starting or resuming a project conversation and allowing Codex to run ordinary read-only project inspection commands can make t…

Trump administration accuses China of dodging tariffs

“It’s big money, but there’s also a very human cost here,” said Peter Navarro, senior counsel for trade and manufacturing.

Let's Try to Improve How Conversational AI Responds to Prematurely Submitted Inputs!

From Response Generation to Response Readiness Hello everyone, I would like to share another thought following my previous post. Conversational AI research has largely focused on how to generate a high-quality response . But there is a more fundamental question: Is the AI ready to respond yet? Consider a user who accidentally presses Enter while still composing a message. The AI may generate a perfectly reasonable answer to the text it received. Yet, from the user’s perspective, that answer may be completely inappropriate — because the user was not finished speaking . This suggests that conversational AI needs a capability beyond response generation: the ability to recognize whether the user’s input has reached a state in which responding is appropriate. In other words, the AI should distinguish between: incomplete input, complete text, complete user intent, and input that is actually ready for a conversational response . This leads to a different design question: When should a conversational AI respond? Rather than treating this solely as a UI problem, I believe it should be considered an architectural problem in conversational AI . A high-quality conversational AI should not only know how to answer . It should also know when to answer — and when not to answer yet. This may represent an important shift: From Response Generation → to Response Readiness. So, should the next generation of conversational AI be designed not only to generate better answers, but also to recognize…

Netflix is closing two game studios

Netflix plans to shut down two of its gaming studios, as reported by Game File and Variety, as it makes a bigger shift toward party games and titles streamed to TVs. One of the studios being shut down is Night School Studio, creators of the Oxenfree series. Netflix bought Night School in 2021, and it […]

Neue KI-Modelle: Gemini 3.7 Flash und DeepSeek V4-Pro im Vergleich

Google bringt Gemini 3.7 Flash, OpenAI beschleunigt GPT-5.6 Sol mit Ultrafast, DeepSeek V4-Pro kommt mit Zeitzonen-Tarif.

Give Google the boot by building your own search engine

Got $10 and a gig of storage space? Marlin lets you weight your crawl toward the parts of the web you actually care about

I tried the new ChatGPT Desktop App for Linux - but I'll stick to my browser for now

This preview release of ChatGPT Desktop for Linux supports Ubuntu, Debian, and Fedora is here.

Finding an apartment in Manhattan is getting even harder, as listings plummet and rents reach $5,000

Median Manhattan rents jumped to $5,000, a new report found. New apartment listings, meanwhile, have nearly disappeared.

Millions across Europe swelter through a new wave of extreme temperatures

Repeat heatwaves and drought have fuelled a historically hot summer of wildfires, and contributed to the deaths of thousands of people.

Building a real-time AI RPG with evolving narratives using LLMs

I am working on something similar for a while. Have GitHub pages describing the project (not yet published). I am unable to paste here project pages describing the project in detail (GitHub static pages) So, the pipeline is getting quite a complexity using orchestrated main pipeline, RAG for lore, history where I use conversation semantic search, memories, etc. Tiered memory system with gradual compression, subagents to act on behalf of active NPCs, weather simulation, map support and detailed scene etc. tracking. Works reasonably well, but it is quite token demanding (using ollama cloud mostly) and lazy (1 turn about 60-90 seconds as of now)

Using functional AI to automate document workflows

A recent study conducted by Nitro found that 75-95% of the employees and executives surveyed use AI for document processing—including data extraction, PDF tasks, and contract summaries. However, when these individuals don’t have access to the right kind of AI tools, they report turning to unapproved—or shadow IT—solutions to speed up workflows, which creates security and compliance risk. Read the report To reinforce the importance of providing teams with the right AI tool for the right job, let’s look at the difference between chatbots and functional AI in terms of automating document workflows. Chatbots are great for ad hoc tasks that follow a pre-programmed set of actions, but they aren’t designed to enforce consistent rules for formatting, redaction, or compliance, or to extract data hidden deep in document tables, images, or free text . Unlike chatbots, functional AI can physically execute redaction, conversion, and data extraction tasks directly within business processes and systems, rather than simply responding to prompts. This guide explains why scaling document workflows requires both conversational AI to answer common questions and functional AI to perform repeatable tasks on a high volume of documents with consistency, control, and predictable cost. Chatbots vs. functional AI: What’s the difference? Chatbot AI and functional AI play distinct roles in document workflows: AI-assisted chatbots answer questions and help users understand documents through conversation.…

Manual vs. AI-powered PDF redaction: protecting sensitive data in 2026

Research shows that humans play a role in 60% of breaches that expose sensitive data. That “role” often involves an employee falling for a phishing scam or using PASSWORD for their login credentials, but data exposure can also be a result of how your business redacts sensitive and personally identifiable information (PII) in your documents. Historically, manual, “black-box” redaction was considered best-practice, but this approach only obscures data, it doesn’t permanently remove it. As regulations governing data security get stricter and AI-powered redaction solutions become more accessible, organizations—especially those in highly regulated industries—are re-evaluating their PDF redaction solutions. How manual PDF redaction is different from AI-powered PDF redaction The primary difference between manual and AI-powered PDF redaction is who (or what) you rely on to do the heavy lifting. Manual redaction defined Manual redaction is a human-driven process where individuals visually scan text, select content, and apply black boxes or remove the text before sharing or storing the file. Manual redaction is only as effective as the reviewer, which makes outcomes highly variable, especially under time pressure or high document volume. AI-powered redaction defined AI-powered redaction uses machine learning and natural language processing (NLP) to automatically detect and remove sensitive information from documents. The system is trained to recognize patterns, language cues, and cont…

Professional PDF solutions for teams: scale without breaking your budget

Why do so many PDF editing and eSignature tools fail to scale across teams? Three persistent problems stand out: Unpredictable pricing that becomes more expensive as usage increases Limited access to advanced PDF features Difficulty automating document workflows across teams The good news is that even teams that process a high volume of documents can reduce their costs and get more value from their PDF solutions by selecting solutions with built-in AI, predictable pricing, flexible licensing, and scalable, automated document workflows. In this article, we’ll cover: Why you’re paying too much for too few features How AI makes PDF editing , eSigning , and redaction tools accessible to everyone on your team What to look for in a scalable, cost-efficient PDF solution Why many incumbent PDF tools are expensive and inefficient at scale Many PDF tools were designed for individuals or small teams rather than scaling teams. As organizations expand usage, these tools become harder to manage, more expensive to maintain, and less effective at supporting high-volume document workflows. Here are three ways PDF tools can impact your budget and your team’s productivity: Unpredictable pricing increases total cost of ownership If your PDF solutions rely on per-user licensing, feature-based tiers, or usage caps that trigger additional fees, as demand for the tool increases, your costs become harder to forecast and control. Limited access to features reduces productivity When full PDF editing f…

Docusign alternatives for teams: what to look for (and what to avoid)

Choose Nitro Sign for predictable pricing & fewer surprises Free Trial Docusign may be the most widely used eSignature platform, but many teams run into the same issue as they scale—the more documents they send, the harder it is to predict what they’ll actually pay. The risk of high overage costs paired with complex pricing structure is driving teams to look for an alternative to Docusign in 2026. This buyer’s guide breaks down why those challenges matter, what to look for in an eSignature solution, and how leading Docusign alternatives—like Nitro Sign —compare for teams that send a high volume of documents but want simpler pricing and fewer surprises. Nitro Why some teams are looking for alternatives to Docusign Three trends drive teams to look beyond Docusign: Bills that increase unpredictably as usage grows Difficulty understanding what features are included vs. which are paid add-ons Pricing models that don’t scale well with high document volume Here’s a more detailed look at each of these challenges. Docusign bills become unpredictable as usage grows Many organizations encounter unexpected costs as usage increases. Envelope limits, counting methods, and per-transaction pricing can quickly turn a straightforward Docusign subscription into an unpredictable expense. Impacts may include: Budget overruns tied directly to business growth Paying for documents that are sent but never completed Difficulty forecasting total cost at scale Docusign pricing doesn’t clearly indicate…

Nitro Smart Redact: the complete guide to automated AI redaction

Get Smart Redact and Protect Sensitive Data Learn more Too many businesses in highly regulated industries—such as healthcare, government, legal services, and insurance—still rely on manual, “black-box” redaction workflows. This approach may obscure sensitive information, but it doesn’t permanently remove it, which can lead to compliance violations, potential litigation, or regulatory fines. Nitro Smart Redact is an AI-powered solution that removes sensitive data from documents with permanent, untraceable redactions. It combines automated PII detection to surface regulated data along with manual controls for sensitive business information, allowing teams to flag content, automatically redact documents, and manually customize reviews that need human oversight. In this guide we explore how Smart Redact: Simplifies permanent, irreversible redaction Identifies sensitive structured and unstructured data Provides enterprise-grade security Reduces turnaround times with automation Minimizes human error How is Smart Redact different from other automated AI redaction solutions? Pattern-based matching tools, like those used in Adobe Acrobat, use predefined formats and keyword matching to locate sensitive data. This approach often misses information in free-form text, like “John lives on Main Street” or “her social ends in 5678.” Smart Redact catches it because it understands context. AI-only platforms—like Redactable—offer automation, but they require teams to maintain a separate redact…

Adobe Acrobat alternatives for teams in 2026

Try Nitro PDF today Learn more If your team is re-evaluating Adobe Acrobat’s place in your 2026 tech stack, the question isn’t just what the software can do, but whether it aligns with your evolving budget and productivity goals. Including Acrobat in your stack is becoming increasingly expensive and difficult to manage due to: Frequent price increases Overlapping licenses that add overhead Underutilized features that waste resources As a result, the conversation is shifting away from “How do we navigate rising costs, complexity, and AI compliance requirements?” to, simply, “ Is there a better alternative to Adobe Acrobat? ” In this guide, we break down: How to stop overpaying for licenses that aren’t being used How to select a PDF editing solution that does exactly what you need it to do Why Nitro PDF is emerging as a leading choice for organizations focused on experience, value, and scalability for IT and procurement-led buying decisions Nitro Why teams are looking for Adobe Acrobat alternatives in 2026 For organizations that use Adobe Acrobat, frustrations tend to build over time—and they usually come down to the same three things: Rising costs that keep climbing What starts as a manageable per-seat subscription becomes harder to justify as Adobe raises prices and layers in paid add-ons. The AI tools, advanced features, and admin tools that should be standard all require extra cost or higher priced plans. Licenses that are hard to manage Adding or removing users, adjusting…

Verified developer blocked from plugin creation

Hi, I keep getting “You need a verified developer identity before you can create or upload a plugin” despite my identity being verified (months ago). Support claims that plugin creation/upload is only available to Business and Enterprise plans but I keep getting the same error even after upgrading to the Business plan. I’m the sole member (owner) of my organization so I have all the permissions and Developer Mode is turned on.

Deep-research not working

Hi team — we’re closing this topic because there hasn’t been a new response for a long period of time and the discussion appears inactive. If you still need help, please start a new topic with your current question and any relevant, non-sensitive details. Thanks for being part of the community. Avinash

OpenAI's Computer History Watches Your Mac So You Stop Re-explaining Yourself

ChatGPT's Computer History replaces the screenshot-based Chronicle preview with an event-driven system that tracks your work across apps and websites without ever capturing your screen.

Responses API + Structured Outputs (gpt-5.6-luna): garbage tokens (foreign scripts / leaked reasoning) inside string values right before the closing quote — identical request via Chat Completions is clean

Update, since @BaileyGranam asked: in the end we couldn’t find any way to prevent this from the request side. We tried everything with the prompt, the schema and the reasoning settings and nothing worked, this really is an upstream bug in the Responses API’s constrained decoder. Since we didn’t want to move back to Chat Completions and lose explicit prompt caching, what we did instead was put a guardrail around the Responses path: Deep-trim every string in the parsed object, always. That silently absorbs the mild cases (trailing whitespace or stray punctuation right before the closing quote), which were way more frequent than the severe ones. For the severe cases, a cheap regex over the raw output_text looking for any script that can’t legitimately show up in our output languages (\p{Script=Han}, Hiragana, Katakana, Hangul, Cyrillic, Hebrew, Arabic, etc.) plus control chars and U+FFFD. One detail: be careful not to block scripts you actually need. We had to leave Greek out of the blocklist because “μg” shows up all the time in nutrition text. If the regex fires, we discard the response and retry once via Chat Completions with the same model, prompt and schema, which is the path that always came out clean for us. Every trigger gets logged and the whole guardrail sits behind an env var kill-switch, so the day OpenAI fixes it we just turn it off and that’s it. The downside: when the retry fires you pay the call twice and lose the cache on that request, but at aprox 6% severe in…

Databricks wanted to raise $1B, investors wanted $15B. It settled on $5B at a $190B valuation.

AI is expensive, Ali Ghodsi tells TechCrunch. With so many investors wanting into his latest round, he said yes to more than planned.

Magyar casts doubt on Hungary-Russia nuclear power deal

Viktor Orbán once hailed the project as the “deal of the century.”

sqlite-utils 4.2

Release: sqlite-utils 4.2 Lots of improvements in this one relating to the table.transform() feature , which adds support for complex alter table operations by creating a fresh table, copying across the data and then dropping and replacing the old one. transform() now preserves a much larger array of edge-case schema definitions, including check constraints, unique constraints and even comments describing the columns. There are also new introspection properties for check constraints, and a whole lot of other smaller changes. Includes contributions from Bunlong Heng , ethanhawkes-gif , Rami Abdelrazzaq , nyxst4ck , and ikatyal2110 . (It later turned out 4.2 had a crashing bug , fixed in 4.2.1 .) Tags: releases , sqlite , sqlite-utils

This school-friendly laptop from HP is $300 off

With memory prices still high and showing no signs of dropping, we’re always happy to find a good deal on a budget-friendly system with an adequate amount of RAM. Best Buy has the HP OmniBook X Flip discounted by $300, bringing the cost down to $699.99. This model is equipped with an Intel Core Ultra […]

Novak Djokovic trains hard as he returns to Cincinnati after three years

Two injured after train derails in southeast England, emergency services say

The ambulance service said there were around 150 passengers on board the train which departed from London Victoria station and veered off the tracks shortly before 4:00 pm local time.

Vietnam discovers more than a tonne of concealed ivory illegally shipped from Africa

Most trade in ivory from elephants and other endangered animals is banned by international agreement under CITES, a treaty to which almost all UN member countries are parties.

Openai api credits grant isssue can i still use those free credits which are still available

Hi there - we’re closing this topic because there hasn’t been a new response for a long period of time and the discussion appears inactive. If you still need help, please reach out to support@openai.com . Thanks for being part of the community. Avinash

Aryna Sabalenka makes history as first Belarusian to grace Time cover

Checking ANCOVA parallel slopes assumption when no degrees of freedom remain for error estimate?

Context I have the following dataset corresponding to a pilot experiment, with two treatment factors and a covariate. Factor A has 3 levels, and factor B has 2 levels. The total number of independent observations is 12. run contains the run order, trtmt corresponds to the treatment combination AB , rate is the covariate, and absorb is the response. run trtmt AB rate absorb 1 1 2 12 1.780 0.7355 2 2 4 22 1.867 0.2828 3 3 2 12 2.136 0.3884 4 4 1 11 1.952 0.6777 5 5 5 31 1.800 0.4364 6 6 1 11 2.000 0.6116 7 7 4 22 2.071 0.2929 8 8 6 32 1.951 0.6465 9 9 3 21 2.273 0.2525 10 10 3 21 2.250 0.2727 11 11 6 32 2.075 0.6707 12 12 5 31 2.158 0.3313 The resource I am using for self-study instructs the reader to plot absorb vs. rate using trtmt as the plotting symbol to see if ANCOVA is appropriate, and discuss whether treatment effects may exist. To me, the plot indicates that ANCOVA is a plausible model, though two treatment combinations seem like they may violate the parallel slopes assumption. We are then instructed to fit a one-way ANCOVA model to the data, and plot residuals against the covariate, run order, predicted values, and normal scores to check model assumptions. I omit these here, but the plots indicate model assumptions are reasonably satisfied. Where I'm stuck Normally, having checked model assumptions, I'd proceed to check the parallel slopes assumption by fitting a model with an interaction effect between the treatment factor and the covariate, then doing an ANOVA comp…