The Trump administration is reportedly considering an executive order to create a working group that could review advanced AI models before public release. The shift follows concerns over Anthropic's powerful Mythos model and its cyber capabilities, with officials weighing whether the government should get early access to frontier models without necessarily blocking their release. The New York Times reports: In meetings last week, White House officials told executives from Anthropic, Google and OpenAI about some of those plans, people briefed on the conversations said. The working group is likely to consider a number of oversight approaches, officials said. But a review process could be similar to one being developed in Britain, which has assigned several government bodies to ensure that A.I. models meet certain safety standards, people in the tech industry and the administration said.
The discussions signal a stark reversal in the Trump administration's approach to A.I. Since returning to office last year, Mr. Trump has been a major booster of the technology, which he has said is vital to winning the geopolitical contest against China. Among other moves, he swiftly rolled back a Biden administration regulatory process that asked A.I. developers to perform safety evaluations and report on A.I. models with potential military applications. "We're going to make this industry absolutely the top, because right now it's a beautiful baby that's born," Mr. Trump said of A.I. at an event in July. "We have to grow that baby and let that baby thrive. We can't stop it. We can't stop it with politics. We can't stop it with foolish rules and even stupid rules." Mr. Trump left room for some rules, but he added that "they have to be more brilliant than even the technology itself."
The White House wants to avoid any political repercussions if a devastating A.I.-enabled cyberattack were to occur, people in the tech industry and the administration said. The administration is also evaluating whether new A.I. models could yield cyber-capabilities that could be useful to the Pentagon and U.S. intelligence agencies, they said. To get ahead of models like Mythos, some officials are pushing for a review system that would give the government first access to A.I. models, but that would not block their release, people briefed on the talks said.
An anonymous reader quotes a report from 404 Media: A new, bipartisan bill introduced (PDF) by Democratic Senator of California Adam Schiff and endorsed by the biggest AI developers in the world -- including OpenAI, Google, and Microsoft -- would change the K-12 curriculum to shoehorn in "AI literacy," something that young people and teachers alike already hate in schools. The Literacy in Future Technologies Artificial Intelligence, or LIFT AI Act, would empower the new director of the National Science Foundation (NSF) to make grant awards "on a merit-reviewed, competitive basis to institutions of higher education or nonprofit organizations (or a consortium thereof) to support research activities to develop educational curricula, instructional material, teacher professional development, and evaluation methods for AI literacy at the K-12 level," the bill says.
It defines AI literacy as using AI; specifically, "having the age-appropriate knowledge and ability to use artificial intelligence effectively, to critically interpret outputs, to solve problems in an AI-enabled world, and to mitigate potential risks." The bill is endorsed by the American Federation of Teachers, Google, OpenAI, Information Technology Industry Council, Software & Information Industry Association, Microsoft, and HP Inc. [...] The grant would support "AI literacy evaluation tools and resources for educators assessing proficiency in AI literacy," according to the bill. It would also fund "professional development courses and experiences in AI literacy," and the development of "hands-on learning tools to assist in developing and improving AI literacy." Most importantly for real-world implications, it would fund changing the existing curriculum "to incorporate AI literacy where appropriate, including responsible use of AI in learning."
An anonymous reader quotes a report from Bloomberg's Ashley Carman: Welcome to the modern era of podcasting in which thousands of new shows are released into the world every day with a sizable portion likely being AI-generated. Figuring out exactly which ones fall into that growing category is becoming more difficult just as the industry is starting to take this issue seriously. In only the past month or so, Amazon launched a feature that explains a product by generating a quasi-podcast, complete with co-hosts talking to each other and taking questions from users. Shout out to Business Insider reporter Katie Notopoulos for spotting this (and, naturally, demoing it with an adult diaper rash-cream). Not long ago, Nicholas Thompson, chief executive officer of the Atlantic, noted "podslop" dominated his Spotify search results when he typed in the word "Sora." This was around the time that OpenAI shut down its user-generated, AI-content-only app.
[...] All of which raises some big, difficult questions. For one, what should the listening platforms do about this incursion? As of right now, Apple Podcasts requires creators who generated a "material portion" of their show using AI to disclose it. The platform also bans misleading or deceptive content. Spotify hasn't published any specific guidelines around AI, though it maintains general rules around dangerous and misleading content. Where this conversation gets even trickier is when it comes to money. Many of these podcasts are hosted on at least one free service that allows programs to opt into their ad marketplace with zero barrier to entry, meaning these shows (and the hosting service) profit off every listen or download. Spreaker, a company owned by iHeartMedia, is the primary one to watch here. Though it tells users to disclose when they rely on AI, it still allows those shows to opt into its programmatic ad marketplace, which pays creators 60% of the revenue generated by the ads placed in their shows. It stands to reason that most of these thousands of shows don't reach many people. But in the aggregate, the ears and dollars could add up. Are the advertisers on board with being next to AI-generated content, some of which might be deemed "slop?" There's also the question of how to define "slop." Jackson of the Podcast Index and his co-host Adam Curry treat it as something listeners simply know when they hear it, while Alberto Betella, co-founder of RSS.com, defines it as "fully automated content with no human review."
Jeanine Wright, co-founder of Inception Point, rejects the debate altogether: "The people still talking about slop are still making 6-7 jokes," she said. "It's still yesterday's conversation."
Le procès entre Elon Musk et Sam Altman, qui pourrait bouleverser l'avenir d'OpenAI, va être diffusé en direct à partir de sa seconde semaine. Problème : le tribunal n'autorise qu'un flux audio. Il sera impossible de le regarder en vidéo.
A Wall Street Journal opinion piece warns of "a troubling trend" in AI's growth. "Rather than selling software, some AI companies are paying their partners to use it."
It cites OpenAI's $1.5 billion joint venture with private-equity firms, Anthropic's $200 million contribution to a private-equity firm joint venture, and Google's $750 million subsidization of Gemini's adoption by consulting firms. "These agreements muddy the distinction between a company's sound growth trajectory and artificial financial engineering."
[T]he scale and structure of the recent AI deals go beyond standard incentive mechanisms... When a seller pays customers to buy its products, it is unclear if its revenue growth reflects vibrant demand or a willingness to accept subsidies.
Slashdot reader destinyland writes:
This warning comes from a prominent figure in the investing community. For six years Robert Pozen was chairman of America's oldest mutual fund company, after five years at Fidelity. An advocate for corporate governance, he's currently a lecturer at MIT's business school (and the author of the book Remote Inc.: How to Thrive at Work...Wherever You Are). "As AI companies prepare initial public offerings, investors should scrutinize their numbers closely," Pozner writes, warning about "time-limited financial support".
"In evaluating AI sales figures, analysts should consider the distorted incentives that the recent financing deals create," writes Pozner:
Private-equity firms, enticed by promised returns, might demand rapid rollouts of AI products, rather than ensuring their orderly and safe development. Portfolio companies of private-equity firms may embrace AI tools not because they are needed but because adoption is mandated by their owners. Consultants may favor one set of AI models based on the subsidy instead of the merits.
If guarantees and subsidies are major factors in the rapid adoption of AI tools, investors should be skeptical of AI companies' revenue projections. Many of their customers enticed by consultants will stop paying full price when the financial incentives are gone. Many of the portfolio companies of private-equity firms could back away from selected AI tools once these joint ventures expire. The challenge with evaluating these AI financing deals is the lack of transparency. At present, AI vendors don't separate revenue driven by subsidies or joint ventures from standard sales.
The lesson from the telecom debacle is that financial engineering can obscure, for years, the difference between real customer demand and demand driven by incentives. When AI companies begin to finance their own product distribution, guaranteeing returns to investors and subsidizing sales, it's a signal for investors to dig deeper.
Investing in an AI company? Ask what percentage of enterprise revenue is coming from subsidized channels or joint ventures, Pozner suggests. And the renewal/retention rate for customers not supported by subsidies or joint ventures...
Elle s’est fait attendre, mais la voici maintenant disponible sur le marché français. Suzuki présente sa première voiture 100% électrique. Après une précédente rencontre dans un environnement contrôlé, nous l’avons conduite sur les jolies routes de la campagne bourguignonne, idéales pour se faire une bonne idée de ce que vaut l’e-Vitara.
Comme le concept-car
On a déjà à peu près tout dit sur le physique de la toute nouvelle Suzuki e-Vitara. Son design devait être en partie compatible avec celui de son cousin, l’Urban Cruiser. Pour faciliter son intégration dans le catalogue, le constructeur japonais s’est appuyé sur son best-seller, le Vitara, qui continue d’exister à ses côtés pour le moment. Il ressemble quasiment trait pour trait au concept-car qui l’a précédé. Il paraît à la fois robuste et plutôt moderne. Avec ses épaulements, il semble vouloir bousculer les autres sur la route.
Ce SUV a une garde au sol relativement élevée, qui lui permet non seulement de franchir les trottoirs, mais aussi de s’aventurer parfois sur des chemins. Car, tout Suzuki qu’il est, il a droit aussi à une version à transmission intégrale, avec un moteur sur le train arrière sur la version AllGrip. Il a son petit look et, il faut bien l’avouer, il donne un sacré coup de vieux à son frère thermique, qui commence véritablement à accuser le poids des ans. La génération de ce dernier existe depuis 2015 et s’est, au passage, écoulée à plus de 54 000 exemplaires. On ne sait pas si c’est l’objectif que se fixe la marque pour l’e-Vitara.
Un intérieur de voiture, pas de vaisseau spatial
À bord, on retrouve cet intérieur que nous avions découvert l’an passé. Oui, Suzuki a pris son temps pour ajuster sa production dans son usine indienne. D’ailleurs, il précise que la pleine cadence n’interviendra pas avant juin. En tout cas, on évolue dans un environnement qui, ici aussi, marque l’écart avec l’autre Vitara. L’atmosphère s’avère plutôt moderne, avec une instrumentation 100% numérique. Quelques revêtements moussés rendent l’ambiance un peu plus chaleureuse. Pas d’écran XXL (Apple CarPlay/Android Auto embarqués), mais disons qu’il est à la bonne taille pour ne pas avoir à faire de grands gestes au moment de le manipuler. Un chargeur à induction, une prise USB-C et USB-A sont présents à l’avant.
Aussi bête que cela puisse paraître, Suzuki a agencé la plupart des commandes comme auparavant dans la plupart des voitures, avec des boutons physiques pour la gestion de la climatisation en premier lieu, et une molette rotative pour le son. Basique ? Loin s’en faut dans l’industrie automobile actuelle. On comprend assez rapidement où se trouvent la plupart des fonctions. On se sent bien à l’avant, et pas trop mal à l’arrière en ce qui concerne l’espace aux jambes. Méfiance tout de même, pour les grands gabarits, au niveau de la garde au toit. Le coffre dépasse à peine 300 litres, mais offre un plancher plat. On peut ajuster sa contenance grâce aux sièges coulissants sur 16 cm.
Temps de recharge long en DC
Après nos premiers tours de roues, on constate une régénération manifestement réglée sur un niveau intermédiaire. On cherche alors à la modifier. On se rappelle alors, lors d’une prise de contact avec un prototype l’an passé, avoir recommandé aux ingénieurs japonais un accès rapide à ce réglage. Malheureusement, ce n’est pas le cas : il faut passer par plusieurs étapes sur l’écran, et surtout être à l’arrêt pour choisir l’un des trois niveaux. Dommage ! Pour le reste, on profite, sur notre version deux roues motrices, d’un moteur de 128 kW (174 ch) couplé à une batterie de 61 kWh (LFP). La voiture ne manque pas d’entrain ; globalement, elle se comporte bien dans toutes les situations de conduite, que ce soit en ville ou sur route.
Néanmoins, on remarque assez rapidement que l’on doit utiliser le mode ECO et fluidifier sa conduite si l’on veut maintenir une consommation raisonnable. Cela se fait sans grande difficulté, mais on voit nettement la différence si on la mène tambour battant. Le chiffre WLTP dépassant les 400 km nous paraît plutôt optimiste. Cependant, il est probable qu’en faisant attention, elle s’approche des 300 km. Avec son chargeur embarqué de 11 kW, on peut espérer une recharge complète en un peu plus de 5 heures. Suzuki refuse de communiquer sur la puissance en courant continu, mais le constructeur annonce un temps de 45 minutes pour passer de 10 à 80%. Il va clairement falloir que cette donnée s’améliore par la suite.
Bon comportement routier
Même si la puissance paraît faible, notez que l’on peut préconditionner manuellement la batterie. Nul besoin, comme chez certains concurrents, de devoir obligatoirement entrer une destination dans le GPS. En tout cas, on ne passe pas un mauvais moment à son volant. La voiture se montre plutôt bien amortie, ce qui sert assez bien le confort. Son couple confortable lui assure notamment des reprises de bon aloi. On doit aussi son bon comportement routier à ses suspensions arrière multibras. Soyons clairs : elle n’a pas pour autant des accents de compacte sportive.
En se perdant sur quelques chemins ici et là, on apprécie la garde au sol de 18 cm. On se sentirait encore un peu plus à l’aise avec l’e-axle associé à la version AllGrip. Mais honnêtement, par temps sec, cela ne nous manque pas outre mesure. L’insonorisation à haute vitesse aurait mérité un peu plus de soin. Même si cela ne s’avère pas trop gênant, on l’a tout de même remarqué. On oublie aussi son poids, proche des deux tonnes, qui ne se fait pas trop ressentir. Il se rappelle parfois à nous lorsque l’on rebondit sur une imperfection du bitume que l’on n’aurait pas anticipée.
Un rare SUV compact électrique 4X4
D’entrée de jeu, se sachant désavantagé en Europe à cause de son assemblage en Inde, l’e-Vitara bénéficie d’une ristourne maison de 4 000 € à son lancement. Ses tarifs vont de 32 500 à 40 800 € hors réduction constructeur. La version AllGrip à transmission intégrale se distingue comme l’une des rares 4×4 sur le segment. Au-delà de la garantie classique de 3 ans, elle peut grimper jusqu’à 10 ans si l’entretien est réalisé dans le réseau selon les préconisations. D’ailleurs, n’importe quelle Suzuki déjà vendue est éligible à cette garantie étendue, après un diagnostic de la marque. L’e-Vitara a mis du temps à arriver avec certaines qualités, mais il faudra progresser, notamment sur la recharge.
"Vibe coding just got a whole lot more adorable," writes Engadget:
OpenAI introduced AI-generated pets to the Codex app, its agentic tool that helps with coding. These "optional animated companions" don't do any coding themselves, but serve as a floating overlay that can tell you what Codex is working on, notify you when Codex completes a task or whether it needs your input on something. The new feature lets developers see Codex's active thread, without having to switch away from your current open app.
"The feature ships with eight built-in variations — including a cat and dog," reports Mashable. "But the more interesting play is the custom pet creator."
Users can prompt Codex directly to generate their own companion, then share it online. A quick scroll through the homepage reveals the community has already gotten to work. Current creations include Goku, Patrick Star, Microsoft's long-retired Clippy, OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei, and — naturally — a goblin.
There's also Grogu, Dobby, a tiny Bob Rossi, and a "Doge-style Shiba Inu dog"...
The Associated Press reports:
On a March afternoon, artificial intelligence detected something resembling smoke on a camera feed from Arizona's Coconino National Forest. Human analysts verified it wasn't a cloud or dust, then alerted the state's forest service and largest electric utility. One of dozens of AI cameras installed for the utility Arizona Public Service had spotted early signs of what came to be known as the Diamond Fire. Firefighters raced to the scene and contained the blaze before it grew past 7 acres (2.8 hectares).
As record-breaking heat and an abysmal snowpack raise concerns about severe wildfires, states across the fire-prone West are adding AI to their wildfire detection toolbox, banking on the technology to help save lives and property. Arizona Public Service has nearly 40 active AI smoke-detection cameras and plans to have 71 by summer's end, and the state's fire agency has deployed seven of its own. Another utility, Xcel Energy in Colorado, has installed 126 and aims to have cameras in seven of the eight states it serves by year's end... ALERTCalifornia is a network of some 1,240 AI-enabled cameras across the Golden State that work similar to the system in Arizona....
Pano AI, whose technology combines high-definition camera feeds, satellite data and AI monitoring, has seen a growing interest in its cameras since launching in 2020. They've been deployed in Australia, Canada and 17 U.S. states, including Oregon, Washington and Texas... Last year, its technology detected 725 wildfires in the U.S., the company said... Cindy Kobold, an Arizona Public Service meteorologist, said the technology notifies them about 45 minutes faster on average than the first 911 call.
The Wall Street Journal reports that OpenAI "recently gave its popular ChatGPT strict instructions. Stop talking about goblins."
Recent models of the artificial-intelligence chatbot have been bringing up the creatures in conversations with users seemingly out of the blue, as well as gremlins, trolls and ogres. The goblin-speak caught the attention of programmers, who are often heavy users of the bot. Barron Roth, a 32-year-old product manager at a tech company, said the bot referred to a flaw in his code as a "classic little goblin." He said he counted more than 20 times it mentioned goblins, without any prompting...
Several users speculated that goblin terminology was how the model characterized itself, in lieu of identifying as a person with a soul. Then OpenAI decided enough was enough. "Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user's query," reads an open source line in ChatGPT's base instructions for its coding assistant.
The Journal calls this "a reminder that even as AI companies tout one advance after another in their technology, they are sometimes baffled by the things their own models do...." While training a "nerdy" personality for their model's customization feature, "We unknowingly gave particularly high rewards for metaphors with creatures," OpenAI explained in a log post. And "From there, the goblins spread."
When we looked, use of "goblin" in ChatGPT had risen by 175% after the launch of GPT-5.1, while "gremlin" had risen by 52%... With GPT-5.4, we and our usersâ noticed an even bigger uptick in references to these creatures... Nerdy accounted for only 2.5% of all ChatGPT responses, but 66.7% of all "goblin" mentions in ChatGPT responses... The rewards were applied only in the Nerdy condition, but reinforcement learning does not guarantee that learned behaviors stay neatly scoped to the condition that produced them. Once a style tic is rewarded, later training can spread or reinforce it elsewhere, especially if those outputs are reused in supervised fine-tuning or preference data.
It all started because the "nerdy" personality's prompt had said "You must undercut pretension through playful use of language. The world is complex and strange, and its strangeness must be acknowledged, analyzed, and enjoyed..." Now OpenAI calls this "a powerful example of how reward signals can shape model behavior in unexpected ways, and how models can learn to generalize rewards in certain situations to unrelated ones."
But "fans of goblins don't have to fear," notes the Wall Street Journal. "OpenAI provided a command in its blog post that would remove its creature-suppressing instructions."
An official in South Africa withdrew a draft of the country's national AI policy, reports a local newspaper, "after it was found the draft policy was compiled using AI, which cited academic articles that were 'fictitious'."
Earlier this month, minister in the Presidency Khumbudzo Ntshavheni announced cabinet had approved the draft policy for public comment. [Ntshavheni] said the policy seeks to strengthen government's ability to regulate and adopt AI responsibly, while fostering innovation, job creation, and skills access.
The article includes this quotes from the country's minister of communications/digital technologies department. "This unacceptable lapse proves why vigilant human oversight over the use of artificial intelligence is critical."
Thanks to Slashdot reader Tokolosh for sharing the article.
In 2016 an online "swarm intelligence" platform generated a correct prediction for the Kentucky Derby — naming all four top finishers in order. (But its 2017 predictions weren't even close.) Slashdot checked in again on how modern AI systems performed in 2023, 2024, and 2025 — but their predictions were still pretty bad. Would AI-generated Derby predictions be any better in 2026?
This year's winner was 24-to-1 longshot "Golden Tempo" — though a lot of oddsmakers had favored a horse named Further Ado (which ultimately only finished 11th). So when USA Today prompted Microsoft Copilot for its own picks for the Kentucky Derby, Copilot also went with Further Ado. (Even worse, it predicted Golden Tempo would come in... 13th.)
Here's how Copilot's picks actually performed...
Further Ado (finished 11th)Chief Wallabee (finished 4th)The Puma (SCRATCHED)Renegade (finished 2nd)Commandment (finished 7th)So Happy (finished 9th)Emerging Market (finished 10th)Danon Bourbon (finished 5th)Potente (finished 12th)Incredibolt (finished 6th)Robusta (finished 14th)Ocelli (finished 3rd)Golden Tempo (finished 1st)Pavlovian (finished 18th)Great White (SCRATCHED)Wonder Dean (finished 8th) Litmus Test (finished 17th)Albus (finished 15th)Six Speed (finished 13th)Intrepido (finished 16th)
Copilot was told to use the latest odds, conditions, and analysis of favorites, best bets, expert picks, previous results and race history with the post positions, according to USA Today.
And meanwhile, Yahoo Sports asked Claude "to simulate the race using the opening odds, draw and potential track conditions. We also asked it to factor in some human predictions."
Like Microsoft Copilot, Claude also picked Further Ado to finish first (though it came in 11th) — and predicted that Golden Tempo (the eventual first-place finisher) would finish 12th.
Further Ado (finished 11th)The Puma (SCRATCHED)Commandment (finished 7th)Chief Wallabee (finished 4th)Renegade (finished 2nd)Emerging Market (finished 10th)So Happy (finished 9th)Incredibolt (finished 6th)Danon Bourbon (finished 5th)Potente (finished 12th)Pavlovian (finished 18th)Golden Tempo (finished 1st) Litmus Test (finished 17th)Albus (finished 15th)Wonder Dean (finished 8th)Six Speed (finished 13th)Intrepido (finished 16th)
"Running a Big Tech company during Silicon Valley's AI mania may not necessarily require fewer workers or cost less," writes the Washington Post:
Amazon, Google and Meta together have roughly the same number of employees now as they did during an industry-wide hiring binge in 2022, company disclosures show. Growing costs for technical workers and related expenses have often outpaced sales recently. The tech giants' big AI bet hasn't yet paid for itself.
That means AI might be killing jobs not through its labor-saving wizardry but by increasing spending so much that CEOs are pressured to find savings, giving them cover to consciously uncouple from their workforces. Marc Andreessen, a prominent start-up investor and a Meta board director, put it bluntly on a recent podcast. Big company layoffs are a fix for overstaffing and changing economic conditions, he said, but AI provides a convenient scapegoat. "Now they all have the silver bullet excuse: 'Ah, it's AI,'" he said...
"Almost every company that does layoffs is blaming AI, whether or not it really is about AI," Sam Altman, CEO of ChatGPT owner OpenAI, said at a March conference when he listed explanations for AI's unpopularity in the United States.
"Recent history suggests Big Tech companies might not be moving toward a future with fewer workers," the article concludes, "but recalibrating to spend the same, or more, on different people and projects."
So in the end, "AI might soon reduce hiring," the article acknowledges, "But the reluctance or inability of the largest tech firms to cut too deeply so far could also show that the path to making a workforce AI-ready — whatever that means — isn't a predictable straight line charting declining headcount."
Slashdot reader joshuark writes: Scientific American reports that a ChatGPT AI has proved a conjecture with a method no human had developed. A 23-year-old student Liam Price just cracked a 60-year-old problem that world-class mathematicians have tried and failed to solve. The new solution that Price got in response to a single prompt to GPT-5.4 Pro was posted on www.erdosproblems.com, a website devoted to the Erds problems. The question Price solved — or prompted ChatGPT to solve—concerns special sets of whole numbers, where no number in the set can be evenly divided by any other...
Price sent it to his occasional collaborator Kevin Barreto, a second-year undergraduate in mathematics at the University of Cambridge. The duo had jump-started the AI-for-Erds craze late last year by prompting a free version of ChatGPT with open problems chosen at random from the Erds problems website. Reviewing Price's message, Barreto realized what they had was special, and experts whom he notified quickly took notice.
ZipNada shares a report from ZDNet: Given the dour headlines as of late concerning the diminishing amounts of entry-level software development jobs, coupled with predictions of applications entirely AI-generated, one could be forgiven for assuming that software developers may soon be an endangered species. However, the data tells a different story. James Bessen, professor at Boston University, has been pushing back for some time against the talk of AI and automation displacing jobs on a mass scale, and lately has been arguing that the roles of software developers are nowhere near extinction.
AI is certainly not killing the software developer, Bessen said in a recent analysis (PDF). AI is taking over software development tasks and boosting productivity and output, but that is not translating into lost jobs, he argued. Instead, the types of software skills sought by companies are changing. "Surprisingly, however, after three years of AI use, software developer jobs have continued to grow robustly, reaching record levels of employment -- 2.5 million in February," Bessen said in the report, citing data from the US Bureau of Labor Statistics. The number of software developers in the US has grown by over 400,000, or 19%, since ChatGPT was introduced in 2022. At that time, the employed software developer population was just under 2.1 million. [...]
The productivity uptick developers are seeing may ultimately be a boost to their professional opportunities, however. "An important and possibly disruptive change is happening, but the common view misunderstands what is going on," Bessen pointed out in his report. "Careful case studies find that AI improves the productivity of software developers -- that is, the software produced per developer -- by 30%, 50%, or more. And the rate of productivity improvement in software development is improving." Tellingly, since 2022, when ChatGPT was introduced, developer productivity has increased noticeably, Bessen continued. "From 2003 to 2022, developer productivity grew at 3.9% per year; but from 2022 through 2025, it grew at 6% per year." [...] A coming flood of new software products, now more likely to be enhanced by AI, will continue to create jobs for developers, Bessen predicted. "Thus, mass unemployment of software developers seems unlikely to happen soon." This doesn't mean the job descriptions of developers or other computer occupations will remain static. AI is shifting and re-inventing these roles, Bessen added.
An anonymous reader quotes a report from Ars Technica: Last month, Anthropic made a big deal about the supposedly outsize cybersecurity threat represented by its Mythos Preview model, leading the company to restrict the initial release to "critical industry partners." But new research from the UK's AI Security Institute (AISI) suggests that OpenAI's GPT-5.5, which launched publicly last week, reached "a similar level of performance on our cyber evaluations" as Mythos Preview, which the group evaluated last month.
Since 2023, the AISI has run a variety of frontier AI models through 95 different Capture the Flag challenges designed to test capabilities on cybersecurity tasks, such as reverse engineering, web exploitation, and cryptography. On the highest-level "Expert" tasks, GPT-5.5 passed an average of 71.4 percent, slightly higher than the 68.6 percent achieved by Mythos Preview (though within the margin of error). In one particularly difficult task that involved building a disassembler to decode a Rust binary, AISI notes that "GPT-5.5 solved the challenge in 10 minutes and 22 seconds with no human assistance at a cost of $1.73" in API calls.
GPT-5.5 also matched Mythos Preview in its progress on "The Last Ones" (TLO), an AISI test range set up to simulate a 32-step data extraction attack on a corporate network. GPT-5.5 succeeded in 3 of 10 attempts on TLO, compared to 2 of 10 for Mythos Preview -- no previous model had ever succeeded at the test even once. But GPT-5.5 still fails at AISI's more difficult "Cooling Tower" simulation of an attempted disruption of the control software for a power plant, as every previously tested AI model also has. The new results for GPT-5.5 suggest that, when it comes to cybersecurity risk, Mythos Preview was likely not "a breakthrough specific to one model" but rather "a byproduct of more general improvements in long-horizon autonomy, reasoning, and coding," AISI writes.
Et si l’intelligence artificielle pouvait ressusciter le génie de Jean-Baptiste Poquelin ? À travers le projet Molière Ex Machina, des experts en IA et des universitaires ont entraîné des modèles de langage pour produire une pièce inédite, des costumes aux décors baroques. Après deux ans de développement, le résultat de cette expérimentation sera dévoilé à l'Opéra royal de Versailles les 5 et 6 mai.
An anonymous reader quotes a report from Ars Technica: The system prompt for OpenAI's Codex CLI contains a perplexing and repeated warning for the most recent GPT model to "never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user's query."
The explicit operational warning was made public last week as part of the latest open source code for Codex CLI that OpenAI posted on GitHub. The prohibition is repeated twice in a 3,500-plus word set of "base instructions" for the recently released GPT-5.5, alongside more anodyne reminders not to "use emojis or em dashes unless explicitly instructed" and to "never use destructive commands like 'git reset --hard' or 'git checkout --' unless the user has clearly asked for that operation."
Separate system prompt instructions for earlier models contained in the same JSON file do not contain the specific prohibition against mentioning goblins and other creatures, suggesting OpenAI is fighting a new problem that has popped up in its latest model release. Anecdotal evidence on social media shows some users complaining about GPT's penchant for focusing on goblins in completely unrelated conversations in recent days. Update: OpenAI has published a blog post explaining "where the goblins came from."
In short, a training signal meant to encourage its "Nerdy" personality accidentally rewarded creature-heavy metaphors, causing words like "goblins" and "gremlins" to spread beyond that personality into broader model behavior. OpenAI says it has since retired the Nerdy personality, removed the goblin-friendly reward signal, and filtered creature-word examples from training data to keep the quirk from resurfacing in inappropriate contexts.
Depuis plusieurs jours, les théories se multiplient autour de l’étrange obsession de certains modèles d’OpenAI pour les gobelins, gremlins et autres créatures fantastiques. L’entreprise vient de publier une explication détaillée, et elle apporte un éclairage sur les limites de l’entraînement par renforcement.
Dans les instructions internes de Codex CLI, l’agent de programmation d’OpenAI, une consigne inattendue revient à plusieurs reprises : ne jamais mentionner de gobelins, gremlins, ratons laveurs, trolls, ogres ou pigeons. Cette interdiction, devenue virale, alimente débats et théories en ligne.
Connexion
