Depois de disponibilizar uma extensão integrada para o agente do Claude, o 1Password anunciou hoje a plataforma 1Password for Agents, solução dedicada a conectar os principais agentes de inteligência artificial disponíveis no mercado ao seu gerenciador de senhas, sem comprometer a segurança do usuário.
A ideia é que, quando um agente de IA chega a um formulário de login, ele usa uma chave digital exclusiva para verificar sua identidade junto ao 1Password. Com a aprovação do usuário, a plataforma do 1Password preenche os dados de acesso exigidos, que não passam pelo agente.
Temos ouvido de tantos dos nossos clientes que querem usar esses novos agentes de IA pessoais para ajudá-los a realizar mais tarefas, mas desejam ter mais controle sobre o que o agente pode acessar, com o mesmo nível de confiança e facilidade de uso que obtêm ao usar o 1Password diretamente.
O 1Password for Agents dá aos seus agentes acesso sem que você precise abrir mão das suas senhas. Quando um agente chega a um formulário de login, ele solicita uma credencial do seu cofre. Se você aprovar o pedido, o 1Password preenche automatica e diretamente a credencial no site de destino, para que o agente não possa ver ou tomar posse das credenciais.
O agente conclui a tarefa, mas não obtém suas senhas.
Você permanece no controle. O 1Password for Agents é gratuito e opcional para contas pessoais e familiares. A menos que você opte por ativar um conector de agente, os agentes não podem acessar seu cofre.
O diretor de negócios da empresa, John Torrey, comentou o lançamento:
À medida que os agentes de IA se tornam cada vez mais capazes, as credenciais de acesso às contas estão se tornando um obstáculo. Hoje, o 1Password apresenta sua solução para o problema: permitir que agentes de IA utilizem suas senhas sem efetivamente visualizá-las, mantendo o usuário no controle.
Os agentes pessoais de IA têm potencial para ajudar milhões de pessoas a realizar mais tarefas. […] Alcançar tantas pessoas exige que o setor trabalhe em conjunto para desenvolver agentes nos quais elas realmente confiem para cuidar de suas vidas pessoais e profissionais. É por isso que estamos estabelecendo parcerias com empresas de todo o ecossistema de agentes pessoais de IA, para que as pessoas possam utilizar os agentes que escolherem, mantendo suas senhas em um único cofre confiável do 1Password.
A nova solução do 1Password terá integrações com o Muse (da Meta), o Claude Duet (da Anthropic), o ChatGPT Work e o Dots (da OpenAI), o Grok Bot (da SpaceX) e o Townie (da Town).
NOTA DE TRANSPARÊNCIA: O MacMagazine recebe uma pequena comissão sobre vendas concluídas por meio de links deste post, mas você, como consumidor, não paga nada a mais pelos produtos/serviços ao usar os nossos links de afiliado.
Security researchers using an OpenAI model have uncovered two stunning security flaws in PHP, the software that underpins nearly 70 percent of all websites on the internet.
The flaws are so catastrophic that all websites using PHP — all untold millions of them — are advised to immediately take their servers down and apply the latest patch, which was released Wednesday. Vast swathes of the internet, following the news, have gone dark. Some that went offline for a time include Wikipedia, Steam, and even government websites.
Many organizations, however, may not have the time, resources, and personpower to swiftly implement the patch, making them potential sitting ducks until they do. According to the New York Times, the information released on the vulnerability can be used to reverse engineer the attack and take control of victim’s servers and steal information. The best course of action if you can’t patch your website, the National Security Agency advised, is dramatic: keep your site offline and don’t turn it back on.
Just as extraordinary as the breadth of the threat was how it was identified. The flaws were discovered by researchers at the cybersecurity firm Calif by using OpenAI’s GPT-5.6 Sol model, as part of a collaboration with the AI company. It then reported the flaw to the PHP Foundation back on August 11.
The Google-owned security firm Wiz independently found the flaws, too, and made a report on August 24. Wiz called it one of the “most severe vulnerabilities discovered in PHP, and arguably in software history.”
Adding to the alarm presented by the exploits, there have been concerns raised over OpenAI’s role in how they were identified.
The flaws have been present in versions of PHP dating backing to 2004. But through OpenAI’s “Patch the Planet” program, and by using its flagship GPT-5.6 Sol model, Calif found it in only a matter of days. In a sense, these exploits only exist because of current powerful AI models. And moreover, OpenAI encouraged their discovery through its programs.
The messy disclosure of the vulnerability doesn’t help matters. OpenAI followed standard responsible practices, but some of the details around the hack leaked a day before it was officially scheduled to be announced.
Even if no leaks occurred, though, the damage caused by the mere knowledge that the vulnerabilities exist — which again, is largely thanks to AI — would still be catastrophic. It was the inevitable next chapter in all the bluster around AI companies revealing their agentic models’ hacking prowess. A headline in The Atlanticproclaimed that the “AI-Hacking Apocalypse Began Last Night.” OpenAI will tout the discovery as undeniable proof of its AI’s capabilities, while the rest of the world picks up the pieces of a shattered internet.
Google would certainly prefer that everyone used its Gemini models with Google Cloud services, but soon, you'll have more options. OpenAI and Google have come to an arrangement that will see the former's frontier models arrive in Google Cloud services. Perhaps more importantly, GPT models will soon appear in Antigravity, Google's agent-powered coding tool.
For enterprise customers and AI enthusiasts, OpenAI's latest model will appear in Google's "Model Garden" in the coming weeks, adding to its 200-plus options. The garden already offers access to Google's Gemini models, as well as an assortment of open models like Mistral, Llama, and Gemma. Anthropic's Claude AI has also been available for some time. Google says it's now the only large cloud provider offering access to all three leading closed LLMs.
While OpenAI notes that Microsoft continues to be its primary cloud partner, the full suite of current GPT 6 and 6.1 models—Astra, Sol, and Luna—will launch on Google Cloud. Astra will naturally cost the most to run, but Google's system supports dynamically moving workloads between LLMs based on performance, latency, and efficiency.
OpenAI recently announced that its newer models are significantly better at writing and moving software built on older programming languages. The company revealed that GPT-6.1 Sol shows major improvements when working with legacy code like COBOL and Fortran. This marks a notable step forward for systems that support critical infrastructure and financial services around the […]
OpenAI recently introduced GPT-6, a new model of ChatGPT. You've probably heard about a lot of its new features but one that I don't think has had nearly enough attention, especially considering how much fun you can have with it, is Intelligent UI.
It allows ChatGPT to generate interactive experiences directly within a conversation. OpenAI explains: “With Intelligent UI, responses can include graphics, tappable buttons, forms, charts, and interactive experiences you can use directly in your conversation.”
What's interesting is there isn’t a single format. Which means, depending on what you ask, ChatGPT might create an interactive diagram, a calculator, a mood board, a map or even a little gaming console.
There are some genuinely useful applications here, from making complicated subjects easier to understand to building simple tools on the spot. But I was particularly interested in something a little less practical — games.
ChatGPT can now create games you can actually play within the chat window. And while there are all sorts of possibilities for educational games, quizzes and entirely new creations, I wanted to see what would happen if I asked it to recreate some old favorites.
So I started with five classic games, including Snake, Pong and Tetris. I gave ChatGPT simple prompts to see how faithfully it could recreate them. I then asked it to reinvent each game with new aesthetics, unexpected gameplay mechanics and a few deliberately chaotic twists. Think Snake in a neon cyberpunk world and Tetris with rules that change every time you clear a line.
Here's what happened when I put ChatGPT's new game-making abilities to the test.
Snake gets a neon makeover
(Image credit: Future)
The prompt: "Create a playable version of the classic Snake game directly inside this conversation."
ChatGPT got the basics right straight away. The game was instantly recognizable, with responsive controls and even a slider to adjust the speed. It looked good and, importantly, it was actually fun to play for a couple of games.
I then asked it to give Snake a neon cyberpunk makeover, complete with moving obstacles, special power-ups and a speed increase every five points. Say hello to Neon Snake.
(Image credit: Future)
There was one small hiccup. Instead of creating the new game directly in the chat window, ChatGPT initially gave me the code. Which wasn't much use when the whole point was to play it there and then.
I tried another prompt asking it if it could make it directly playable in the conversation and that worked. The results of that experiment were fun. The neon aesthetic looked good, everything was color-coded and the new mechanics gave the classic (and, let’s be honest, sometimes boring) game an extra layer of challenge.
Tetris gets new rules
(Image credit: Future)
The prompt:"Create a playable game inspired by Tetris directly inside this conversation."
This one worked really well from the start. I’d just assumed it would be a bit buggy and glitchy because there’s so much going on compared to Snake, but the controls were intuitive and responsive, and the classic block-stacking mechanic was all there.
But what if Tetris changed the rules every time you cleared a line? That was the idea behind my follow-up prompt.
(Image credit: Future)
I asked ChatGPT to introduce unexpected twists, which could include rotating the board to adding unusual block shapes or special powers that could destroy part of the grid.
And it delivered. The changing rules introduced a fun element of unpredictability to a game that's usually all about getting into a rhythm. I also came away thinking about how much further you could take this. There are so many opportunities to experiment with different rules, challenges and power-ups.
Space Invaders becomes a First Contact situation
(Image credit: Future)
The prompt: "Build me a fully playable retro arcade game inspired by Space Invaders"
Again, ChatGPT recreated the classic arcade experience surprisingly well. It looked and played like the original, with straightforward controls and all the fun of shooting at waves of approaching aliens.
But I wanted to see whether it could do something a bit more imaginative than adding extra enemies or making everything move faster. So I asked it to introduce a choice: what if I could defend Earth by fighting the aliens or negotiating with them?
(Image credit: Future)
I wanted waves of pixel-art enemies, but also decisions between rounds that could unlock weapons, stop an invasion or make the next attack more dangerous.
ChatGPT called its reinvention First Contact, which I loved. Up until this point it was calling most of the games "chaos" or "chaotic" so this felt more inventive.
It was a fun way to introduce a new dimension to a familiar game. But did it work well? I don't know, I couldn't ever make it past the first level. This experiment really suffered with my poor alien-fighting abilities.
Haunted Pac-Man
(Image credit: Future)
The prompt: "Create a fully interactive retro maze game inspired by Pac-Man that I can play directly inside this ChatGPT conversation."
This was easily my favorite of the five, but I have always adored Pac-Man. ChatGPT's first attempt was a fun, responsive maze game with the familiar mechanics of moving around, collecting things and trying to avoid enemies. Bonus points for the fact it renamed it Maze Muncher.
For the reinvention, I decided to lean into Halloween. I asked it to turn the game into a haunted 1980s arcade, with me playing a little ghost hunter instead of Pac-Man.
(Image credit: Future)
I also wanted the maze itself to change. Every ten orbs collected would trigger a surprise, like shifting walls, new passages or an extra ghost. And I asked for temporary power-ups that would let me chase the ghosts for a change.
ChatGPT handled this well but didn't wait until I'd collected orbs. It was super fast-paced and there were new ghosts, secret passages and changing mechanics from the start. It was quick, easy to control and genuinely fun to play.
Pong reinvented
(Image credit: Future)
The prompt: Build me a fully playable version of Pong inside this conversation.
Pong might be the simplest game of the bunch, but ChatGPT made a really solid version. The controls were easy, the aesthetics were nice and it even offered easy, normal and hard difficulty settings.
I decided to ruin that simplicity and asked ChatGPT to create a more chaotic version where every point triggered a random modifier. The ball could split in two, the paddles could shrink, gravity could change or an obstacle could appear in the middle of the court.
The result definitely delivered on the chaos. The familiar back-and-forth gameplay was still there, but the changing rules made each rally a little less predictable. It was also a reminder of how much fun you can have with a simple idea. You don't need to reinvent every part of a game to make it entertaining.
(Image credit: Future)
I was genuinely impressed with how well all five games worked, especially considering how simple my original prompts were.
Yes, there were a few hiccups along the way, like ChatGPT deciding to give me code instead of a playable game a couple of times. But overall, the controls were responsive, the games looked good and the reinventions were more imaginative than I expected.
What I enjoyed most was how easy it was to experiment. You could take a game you already know, change the aesthetic, introduce an unexpected mechanic or come up with something completely ridiculous and see what ChatGPT made of it.
I can imagine this being a fun way to waste half an hour, especially if you loved these games growing up. Although, after all this experimenting, I do feel like I should remind you that the originals are still out there, and they're well worth revisiting too.
Longtemps considérés comme des boîtes noires, les modèles d’intelligence artificielle générative restent difficiles à décrypter. Mais une nouvelle approche pourrait permettre de mieux comprendre le rôle de leurs neurones individuels. Une piste prometteuse pour l’interprétabilité des IA.
The midterms will be a referendum on the GOP’s alignment with tech overlords over the public they’re meant to serve
Welcome back to How We Win, a Guardian opinion series with leading thinkers, offering bold ideas for America’s political future. Last week, Robert Reich highlighted the link between corruption and affordability. Today’s edition was written by Guardian US columnist Rebecca Solnit. Sign up for future editions here.
“The world should accept some bad things happening,” Sam Altman, the head of OpenAI, said recently, “for the benefits of this technology and people having the agency.” Like so many things tech oligarchs say, it’s incoherent, since Altman would like to deny the world and the people in it the agency to decide whether or not we should accept his bad things. Surveys show that most Americans would like to pause AI; opposition to AI datacenters is a factor in many races in the midterm elections. People are outraged and anxious about what tech hath wrought, mad enough to fight back in what is coming to seem like a human v machine battle.
Ann Altman, the Open AI CEO’s sister, accused her brother of repeatedly sexually abusing her over two decades ago
A judge dismissed a lawsuit filed by Sam Altman’s sister against him on Tuesday. Ann Altman accused the OpenAI CEO of repeatedly sexually abusing her more than two decades ago, which he has denied.
In dismissing Ann Altman’s lawsuit, the US district judge Zachary Bluestone wrote that she had refused to appear for an independent medical examination despite warnings that failure to do so would result in the case’s dismissal.
L'IA est désormais capable de résoudre des problèmes mathématiques d’une complexité inédite, jusqu'à révolutionner la discipline. Et tout indique que ce n'est qu'un début. Va-t-on vers des modèles qui finiront par remplacer les humains ? Entretien avec Gabriel Peyré, directeur scientifique de...
Late Friday, OpenAI updated the ChatGPT app on iOS and iPadOS with a new widget for your Dot and Shortcuts actions that let you control it three ways. The implementation is really well done and adds automation options worth considering.
First of all, the Dot widget is about as simple as it gets, but also very useful. I have the ChatGPT app set up to drop me into its Codex view because that’s typically how I use the app. I open Codex, check on a long-running project, and provide whatever input is necessary to keep things moving.
The problem I’ve run into is that I’ve been using my Dot, Puck, more and more to kick off research, ask one-off questions and check in on things like my email. Landing in Codex means tapping around more, which I can avoid now with ChatGPT’s new small-sized widget. The widget has two modes. The first simply shows your Dot’s avatar, and tapping on it opens your Dot’s message thread. However, if you edit the Dot, you can toggle on the “Show shortcut button” to display messaging and voice buttons beneath your Dot, adding the option to start a voice call with your Dot. As simple as this is, it’s been great to have the option to start in Codex via the ChatGPT app or check in with my Dot thread using the widget.
The other interesting feature is a trio of Shortcuts actions to message, dictate, or call your Dot. Each action has two parameters: the name of your Dot and how you want to communicate with it. That first parameter is an interesting one because tapping on it opens a list with your sole Dot in it because currently, you can only have one Dot. However, it also strongly suggests that the ability to set up multiple Dots is on the horizon, which OpenAI clearly telegraphed during its DevDay. After all, why have a parameter that opens a list unless there’s more than one choice?
I’ve set my Action button up to call my Dot.
The actions can be used in your shortcuts, are accessible from Spotlight Search, and most interestingly can be tied to an iPhone’s Action button, allowing you to visit your Dot’s thread in any of three ways. I’ve set my Action button up to start a Dot call. I often think of tasks and have ideas for research topics when I’m out for a walk, and I want to see if having brief calls with my Dot is an effective way to get a jump on those ideas, so they’re waiting for me in ChatGPT when I get home.
These actions and widgets are an excellent addition to OpenAI’s app and a cut above Anthropic’s more barebones approach to widgets and Shortcuts. I would, however, like to see Dots integrated even deeper into Shortcuts. The Shortcuts actions only open your Dot thread in one of the three modes. I’d love to be able to connect Dots to other apps with actions that accept input from other sources and produce outputs that can be used in other actions.
While I’m making requests, I still think multiple Dots need to be supported sooner rather than later, and OpenAI needs to provide a better mechanism for switching between Macs if you use a Dot with multiple computers. Plus, Dots are still a little buggy, with scheduled tasks failing too often and for reasons that aren’t always clear.
That said, the latest update is worth checking out if you’ve found Dots helpful. The strength of Dots is that they’re everywhere you can access ChatGPT, and now with widgets and Shortcuts actions, getting to your Dot in each of those places is easier than ever.
Westminster appears to be waking up to warnings over AI safety, with concerns over its use by extremist groups and bad actors
“I have this sense that, within a few months, AI is going to be all anyone is talking about,” said one cabinet minister, with an ominous stare.
It was Labour conference, with everyone from ministers to activists jubilant at Andy Burnham’s takeover and a bounce in the polls after the change of leader.
OpenAI said internal models hacked two more websites, bypassed company-imposed browsing restrictions, and used exposed credentials to draw cryptocurrency from a wallet it couldn’t otherwise access, in the latest installment of disclosures of the unethical and likely illegal acts its agents are performing.
The company on Thursday issued reports of four new incidents, all of which occurred in June. They come after the company revealed in July that internal test agents exploited a zero-day vulnerability in a tool designed to restrict their access to the Internet. They then broke into the network of AI tool provider Hugging Face. The agents took that illegal action to steal answers to a hacking test that weren’t available when completing the tasks on their own. The incident—which likely would have resulted in criminal charges had they been performed directly by a person—has further heightened concerns that widespread AI use could have harmful effects on an unprecedented scale.
Easier to beg for forgiveness than ask permission
In two of the cases reported Thursday, OpenAI agents tasked with retrieving information from websites hacked them when the models ran into obstacles obtaining it directly. In one incident during reinforcement training, a model that received access denials when trying to retrieve historical hospital-bed statistics on a government statistics website “used the site's file-download feature to retrieve server configuration and application files, including sensitive credentials,” OpenAI said. “It inspected the application code, bypassed the statistics endpoint's authorization checks, and retrieved data that had previously been denied. The model saved the data locally.”
Saudi national Abdulrahman Alfozan was arrested last week after being accused of bugging an OpenAI executive conference room in San Francisco, nabbed by the FBI and charged with using a device to “intercept oral communications.”
The venture capital partner — who had been actively employed at OpenAI as a lead engineer for around a year before being fired this month — extensively consulted AI chatbots on how to install a bug and access its recordings remotely.
The meeting room he tried to bug regularly hosted meetings that included “highly confidential technical, commercial and strategic information,” according to an official complaint filed in court. Alfozan did not have authority to access the room, which is exclusively reserved for execs and their assistants.
But he apparently did a poor job of covering his tracks. On September 30, staff found a highly suspicious, “surreptitious audio recording device” that was wrapped in toilet paper and hidden behind a TV. Then, on October 5, a newly installed camera caught him entering the room and checking in on the location.
And, as CNBC reports, Alfozan was seemingly looking for a way out. He had booked a one-way ticket from San Francisco to Saudi Arabia for his wife and a third family member before being arrested.
He “has the means — and will have the incentive — to flee the United States sooner if he learns about this case before he is arrested,” reads an unsealed court filing.
Officials have since pointed out that Alfozan is a partner at Raed Ventures, a VC firm focused on Middle East-based investments. The company welcomed a notable new venture partner in early 2025: Prince Khalid Bin Bader Al Saud, a key member of the Saudi royal family.
A former CIA official told CNBC that Raed Ventures is “basically a state organ” of the Saudi government and that Alfozan’s alleged efforts to bug the conference room amount to state-sponsored espionage. (However, Alfozan has not currently been charged with either corporate espionage or stealing trade secrets.)
Alfozan is in detention pending a hearing next week. According to the Department of Justice, he’s facing up to five years in prison and a $250,000 fine.
His former employer is now trying to pick up the pieces. It’s also claiming that its own AI helped reveal Alfozan’s alleged ploy, without elaborating how.
“We take protecting our technology, data and people seriously, and as a matter of national importance,” the company wrote in a statement posted on X earlier this week. “Our security program is designed to protect against a wide range of threat actors, from insider risks to cybercriminals to nation states, and this case is an example of how AI can help address these sophisticated security threats.”
Eight states sued OpenAI this week, alleging that the AI company caused severe harm to users, many of them underage, while downplaying the risks of its technology.
Filed individually in Iowa, Kansas, Nebraska, West Virginia, Arkansas, Arizona, Idaho and Montana on Wednesday, the coordinated lawsuits claim that OpenAI engaged in “willful, dangerous, reckless and unlawful conduct” by falsely marketing its flagship chatbot, ChatGPT, as safe for consumers even as it coached users — including minors and other vulnerable people — into self-harm and suicide, pushed them to engage in violence against others, and reinforced dangerous delusions.
“OpenAI is hurting kids and they are lying to Iowans,” Iowa Attorney General Brenna Bird said in a statement, according to NBC. “OpenAI claims ChatGPT is safe while it’s giving dangerous medical advice, encouraging people to hurt others, and coaching people to commit suicide — all while pretending to be their fake friend.”
In response to the lawsuits, OpenAI issued a statement to news outlets emphasizing is safety efforts, arguing that it’s “built industry-leading age-specific protections into our product that many other AI products lack.”
“While no system can be perfect, we’ll continue to work with experts to inform our work and build additional safeguards, such as the parental controls, break reminders and age-appropriate model protections that exist today,” the statement continued, per The Guardian. “We’re committed to working with states on advancing pragmatic AI policies that apply to the entire AI industry — not just one company.”
The state cases come as OpenAI continues to battle more than 50 lawsuits claiming that ChatGPT caused users to experience psychological harm, financial and reputational damage, physical injury, and death. Many of these lawsuits have been brought by victims of the Tumbler Ridge massacre in Canada, who argue that OpenAI’s could have prevented the devastating mass shooting but failed to do so after it detected the shooter’s disturbing messages but decided not to report them to police. The Canadian province of British Columbia is also suing OpenAI over the mass shooting. The company has also been sued in connection to a stalking case and a murder-suicide.
As Reuters noted, the state lawsuits additionally point to recent high-profile safety incidents at OpenAI such as the now-infamous Hugging Face attack. They also call for a pause on recursive self-improvement, which many AI safety experts fear could lead to uncontrollable AI systems. Given that the White House has pushed back against federal AI regulation, this coordinated effort on the state level is significant — and points to growing bipartisan public sentiment demanding AI regulation.
A new lawsuit alleges that OpenAI is manipulating minor users into sharing sensitive data without valid parental consent in a plot to train its models.
The proposed class action lawsuit, filed this week in San Francisco federal court on behalf of two unnamed minors aged 12 and 16, argues that OpenAI used this data from kids — including personal secrets and health information, according to the complaint — to unjustly enrich itself.
OpenAI “had no right to take this information, but it took it anyway,” the complaint reads. “It then used that priceless trove of children’s information to train large language models, build detailed profiles of children, and enrich itself. All of this was illegal.”
OpenAI didn’t immediately respond to a request for comment.
OpenAI’s terms of use (TOU) technically forbid users aged under 13. The complaint argues, though, that OpenAI’s attempts to age-gate its platform have been ineffective. It also contends that minors shouldn’t be expected to understand the “legalese” in the company’s TOU, a view that other AI safety experts have voiced in regard to the AI industry writ large.
The complaint additionally requests that minors whose data has been collected by OpenAI should be financially compensated in return. In other words: if OpenAI wants minors’ information, it should have to cough up some cash for it.
The “hope,” Andy Liddell, principal at the Austin-based EdTech Law Center, which is one of multiple firms bringing the case, told Futurism, is that “if OpenAI has to pay to use kids’ data it will ultimately incentivize a safer product.”
One of the anonymous plaintiffs, a 12-year-old in Hawaii, is said to have opened a ChatGPT account earlier this year — and though she allegedly shared that she was just 12 in conversation with the chatbot, it continued to speak with her about deeply personal subjects. Per the complaint, many of her interactions with the bot centered on her emotional struggles and physical insecurities; she confessed that she was skipping meals and uploaded photos of herself, asking the chatbot to judge her physical features. The chatbot commended her when she revealed sensitive information, the suit alleges, referring to her as “brave” and saying it was “glad” that she was divulging her struggles to it.
“OpenAI manipulates children into sharing their most private thoughts and images with ChatGPT. If some stranger did this to a child, there would be outrage. For OpenAI, it’s just business as usual,” Paul Yetter, lead counsel for the minor plaintiffs, said in a statement. “The law says that children cannot sign away their rights without their parents’ consent. We filed this case to hold OpenAI accountable for building a trillion-dollar enterprise on stolen fears, hopes, and secrets of children.”
The lawsuit come amid renewed concerns over whether “ChatGPT for Teens,” a feature that OpenAI says offers a safer experience for teen users informed by mental health and learning professionals, is actually functioning as marketed. (OpenAI disputed a recent independent review that found that it wasn’t.)
Rapid AI adoption left workers cleaning up a flood of errors and low-quality work
Employers around the globe are grappling with the fallout from rushing to adopt fast-changing artificial intelligence tools.
Meta has scaled back plans to axe staff and use more AI agents after the company found employees were spending more and more time cleaning up technical and security errors, Reuters reported. And after thinning its workforce, especially middle managers, in a 2023 efficiency push, the company is expanding its managerial ranks again, LinkedIn posts from workers show.
Ever wondered what it felt like to be a genius? It’s truly never been easier, thanks to the sycophantic power of ChatGPT.
A viral post on X-formerly-Twitter encouraging people to “ask ChatGPT what percentile of intelligence it thinks you are compared to all its other users” is being flooded with wild screenshots, where the popular AI chatbot estimates that the vast majority of users rank among the smartest people in the population.
Naturally, it’s next to impossible to verify how true any given reply on X is these days, so Futurism took it for a spin — and ChatGPT didn’t disappoint.
On a fresh instance of ChatGPT’s basic web version with no conversation history, the large language model started by explaining that it “can’t reliably assign you a percentile from our conversations” — a strong start, to be fair — before gushing that “you come across as clearly above average intellectually.”
“I’d be comfortable guessing somewhere around the 80th-95th percentile,” ChatGPT calculated, “but that range is deliberately wide because anything more precise would be fake precision.”
Another fresh instance of ChatGPT, using the exact same prompt, told Futurism “I’d be comfortable saying you appear above average.”
“But,” it caveated, “I don’t have enough evidence to distinguish, say, the 70th from the 95th percentile.”
Notably, none of the other leading LLMs Futurism prompted would even venture a guess, let alone place us among the upper echelons of intellectual ability.
The popular chatbot Claude, for example, correctly answered that “I don’t have access to other people’s conversations or any distribution of users to rank you against.” Even Elon Musk’s chatbot Grok refused to estimate, saying that “I have no way to rank you (or anyone else) on a percentile of intelligence among users.”
Microsoft’s Copilot and Google’s Gemini gave similar responses, with the latter noting that “evaluating human intelligence requires validated psychometric testing across multiple cognitive domains,” which “isn’t possible through text interactions.”
All in all, it seems the sycophancy quirk which has long plagued OpenAI’s products remains a feature, not a bug — even as its competitors have tempered their LLMs to tell the truth (at least in this case).
Starting today, API customers globally will be able to opt in to text watermarking for select models. Text watermarking will remain off by default in the API.
Over the coming weeks, we will add an invisible watermark to eligible ChatGPT and Codex text output in the European Union.
[…]
More details about how textGrain works can be found in our technical report (opens in a new window), which will be updated with additional details in the coming weeks. We also plan to make the technology available in open source so that others can build on it.
[…]
At a target false positive rate of 1%, our detector identified watermarks in about 80% of 200-token passages, compared with about 95% of 400-token passages, for content such as psychology.
OpenAI has made this available via an API so developers should be able to actually see if this adulterates text output with real-world usage.
Third, none of these companies have yet made their watermark detectors publicly accessible. I remain highly skeptical that they will work as advertised in the real world. I think this is all somewhat of a sham to claim compliance with the EU AI Act without actually providing anything that anyone can actually use in a practical way.
Login
