In early April, we shared that more than 3 million developers were using Codex every week. Just two weeks later, that number has grown to more than 4 million.

🔹 **Rich World Knowledge:** Leads all current open models, trailing only Gemini-3.1-Pro.

🔹 **Enhanced Agentic Capabilities:** Open-source SOTA in Agentic Coding benchmarks.

⚠️ Note: deepseek-chat & deepseek-reasoner will be fully retired and inaccessible after Jul 24th, 2026, 15:59 (UTC Time).

🔹 **1M Standard:** 1M context is now the default across all official DeepSeek services.

🔹 **DeepSeek-V4-Flash:** 284B total / 13B active params. Your fast, efficient, and economical choice.

🔹 **DeepSeek-V4-Pro:** 1.6T total / 49B active params. Performance rivaling the world's top closed-source models.

Ubuntu 26.04 LTS provides the strongest foundation for our confidential computing stack. It allows us to deploy a single securely designed image for all our verifiably private AI workloads across Intel, AMD, and NVIDIA hardware, with no platform-specific changes required.

Ubuntu now fully supports RVA23, the baseline standard for RISC-V. This ensures that teams innovating on RISC-V can take full advantage of the platform, including in mixed-architecture environments.

TPM-backed full-disk encryption is now generally available in the Ubuntu installer.

Ubuntu 26.04 LTS is the first LTS to expand the number of memory safe system components. In practice, this means new kernel drivers and subsystems written in Rust, as well as `sudo-rs` and `uutils``coreutils` bringing memory-safe reimplementations of foundational system tools such as `sudo`, `ls`, `cp`, and `mv`.

Canonical Livepatch now extends its rebootless kernel patching capability to Arm64 for the first time.

IgH Master driver brings microsecond-level timing precision natively into the OS, removing a significant integration burden for engineers building motion control systems, robotics platforms, or complex factory automation.

Ubuntu 26.04 LTS is built on Linux 7.0, continuing Canonical's commitment to shipping the latest upstream kernels at the time of release.

With optimized images across AWS, Azure, Google Cloud, IBM Cloud and Oracle Cloud, developers and enterprises can rely on Ubuntu 26.04 LTS for their most demanding public cloud workloads.

The 11th long-term supported release of Ubuntu delivers deep silicon optimization and state-of-the-art security for enterprise workloads.

_Self-reported score with custom Anthropic scaffold._ SWEPro were evaluated with the mini-swe-agent scaffold. However, we use the scores reported by Anthropic for Opus with the max thinking efforts due to frequent timeouts during our evaluation trials.

The depth of recursion becomes a tunable compute axis at inference time, requiring no retraining. A small model, by reading itself, can iterate toward answers that neither it nor any of its workers could reach in a single pass.

Sakana Fugu models are based on our ICLR 2026 papers (**Trinity** and **Conductor**), and we have substantially further improved the methods to increase the performance and user experience

Two variants are available: **Sakana Fugu Mini 🐟**, optimized with latency in mind, and **Sakana Fugu Ultra 🐡**, the full orchestration system, optimized for performance for demanding tasks.

GPQAD | 94.4 | 90.9 | 92.7 | 92.4 | **95.1** | LCBv6 | 90.3 | 92.1 | 92.4 | 90.4 | **93.2** | SWEPro | 48.4 | 51.2 | _53.4_ | 51.3 | **54.2**

Initially, our Sakana Fugu model will be available as an **API**, where it has served as a key internal tool for our own researchers and engineers

Each cell shows how often a given curve fit is not significantly worse than the fit with the best cross-validation accuracy.

We examine whether AI capabilities are accelerating by fitting statistical models to benchmark performance over time, and comparing their predictive accuracies.

Reasoning models show both a one-off jump in performance and a roughly 2-3x faster trend compared to non-reasoning models.

Three of four metrics show strong evidence of acceleration, driven by reasoning models.

Our fourth metric, an index constructed from WeirdML V2 results, showed no sign of acceleration. A single global linear trend fit the data best.

We have been calling this the 'reasoning' / 'non-reasoning' split, but this is not a perfectly clean dichotomy. Several correlated but not strictly identical changes happened over the same few months: scaling inference compute, heavier use of RL in post-training, and models producing reasoning tokens.

The best-performing model across these three metrics was a pair of independent linear trends: one for reasoning models and one for non-reasoning models.

Reasoning models show both a one-off jump in performance and a roughly 2-3x faster trend compared to non-reasoning models.

Three of the four metrics (ECI, log METR 50% time horizon, and a math-focused index we constructed from several math benchmarks) show strong evidence that progress has sped up relative to a global linear trend fit to data from 2023 onward.

Parameters are estimated by unweighted least squares. Time t is measured in years since the first observation in each dataset.

We pre-selected the 6-month horizon as our primary metric, balancing genuine forecasting distance against the limited date range of our data.

The minimum training cutoffs are: ECI (June 2024), METR Time Horizon (January 2024), Combined Math (September 2024), and WeirdML V2 (January 2025).

Our fourth metric, an index constructed from WeirdML V2 results, showed no sign of acceleration. A single global linear trend fit the data best.

We use four AI capability metrics: ECI (Epoch Capabilities Index), METR 50% Time Horizon, Combined Math Index, and WeirdML V2 Index.

Reasoning models show both a one-off jump in performance and a roughly 2-3x faster trend compared to non-reasoning models.

Three of the four metrics (ECI, log METR 50% time horizon, and a math-focused index we constructed from several math benchmarks) show strong evidence that progress has sped up relative to a global linear trend fit to data from 2023 onward.

The three metrics where we find acceleration are concentrated in programming and mathematics. These are areas that labs have explicitly targeted for improvement

Our fourth metric, an index constructed from WeirdML V2 results, showed no sign of acceleration. A single global linear trend fit the data best.

The best-performing model across these three metrics was a pair of independent linear trends: one for reasoning models and one for non-reasoning models.

Reasoning models show both a one-off jump in performance and a roughly 2-3x faster trend compared to non-reasoning models.

Three of four metrics show strong evidence of acceleration, driven by reasoning models.

The three metrics where we find acceleration are concentrated in programming and mathematics.

We select the median-difficulty question from the set with maximum model coverage and standardize it to 0.

We work with the natural logarithm of the time horizon, which puts it on an approximately linear scale.

The minimum training cutoffs are: ECI (June 2024), METR Time Horizon (January 2024), Combined Math (September 2024), and WeirdML V2 (January 2025).

Reasoning models show both a one-off jump in performance and a roughly 2-3x faster trend compared to non-reasoning models.

Three of four metrics show strong evidence of acceleration, driven by reasoning models.

Meta founder and CEO Mark Zuckerberg described superintelligence in a blog post last year

Wedbush Securities analyst Dan Ives said in a report on Thursday.

The layoffs will start on May 20, the company confirmed.

Meta plans to lay off roughly 8,000 employees, or 10% of its workforce

Drug manufacturers pay pharmacy benefit managers rebates above 50% of list price for formulary access.

Google Cloud launched a parallel $750m fund to pay McKinsey, Accenture, and Deloitte to train engineers and co-fund client AI projects.

Structure: $500M OpenAI equity plus $4B from TPG, Bain, Advent, Brookfield, and Goanna form a $10B LLC.

OpenAI pledged $1.5B to a joint venture called DeployCo, guaranteeing private-equity partners a 17% annual return floor over five years.

Amazon is investing $5 billion in Anthropic today, with up to an additional $20 billion in the future

run-rate revenue has now surpassed $30 billion, up from approximately $9 billion at the end of 2025

committing more than $100 billion over the next ten years to AWS technologies

over one million Trainium2 chips to train and serve Claude

over 100,000 customers now run Claude on Amazon Bedrock

up to 5 gigawatts (GW) of capacity for training and deploying Claude

over one million Trainium2 chips to train and serve Claude

run-rate revenue has now surpassed $30 billion, up from approximately $9 billion at the end of 2025

Amazon is investing $5 billion in Anthropic today, with up to an additional $20 billion in the future

more than $100 billion over the next ten years to AWS technologies

over 100,000 customers now run Claude on Amazon Bedrock

up to 5 gigawatts (GW) of capacity for training and deploying Claude

JavaScript is not available. We've detected that JavaScript is disabled in this browser.

JavaScript is not available. We've detected that JavaScript is disabled in this browser.

one of the things that I find really interesting that's not talked about very much is the impacts of nitrogen fixing and the production of artificial fertilizers which contributed to the number one issue which is human population growth

if we exceed 1.5° C, we're likely going to push both the Amazon basin and the top of Coral Reef systems across the threshold if we permanently stay above 1.5.

renewable energy has almost certainly passed a positive tipping point

the AMOC also affects another important climate system, the position of the Inter-tropical Convergence Zone, which is responsible for the monsoon season in India and West Africa

Saisie des suffrages dans ECECA

ECECA : OUTIL DE COLLECTE DES DONNÉES

Point Zero, Vol. I, a book on civilizational development by Boris Sirbey

This way when you use client side routing, you will only receive the _data.json generated at build time and use that to populate the page component with required data (kinda like Next.js does it).

The user variable is shared by everyone who connects to this server.

he represents this in his writing with the point um it's it's the uh everything is a portal to everything else everything is in relationship with everything else there is nothing that is not in relationship with everything uh one point is a doorway to All Points

Different types of leverage points are possible for each infra/exostructureof language. To the extent that each infra/exostructure has its particular way

I believe that we've framed the problem the wrong way. And for me, the big change happened in the late 1970s, when I became aware that there was a battle raging in our westernmost archipelago, what were once called the Queen Charlotte Islands, but are now called Haida Gwaii.

the point of futuring is that you need to connect facts and fictions because that is how this these future Visions become socially performative

Prettier intentionally doesn’t support any kind of global configuration. This is to make sure that when a project is copied to another computer, Prettier’s behavior stays the same. Otherwise, Prettier wouldn’t be able to guarantee that everybody in a team gets the same consistent results.

turn any job – whether in marketing, HR, finance, operations, or beyond – into a climate job.

if we want to achieve transformation what is crucial is that we work across four different levels and that would be Behavior change culture change system change and also work at the level of mindsets

Jordan Hall calls this the Civium, right? Civilization is place-based, and the civium is not place-based. You can still learn.

like it or not Fate has placed the current generation in a position will where it will determine whether we march on the disaster or whether the human species and much other life on Earth can be saved from a terrible Indescribable fate

if Consciousness were to go floating around how the hell do you explain how it can see anything without eyes

To put it bluntly, Web3 and the crypto economy is still largely an ‘exit’ play for financial and coding elites, practicing the arbitrage of nation-states, but without much connections to local communities and resilient production; Similarly, local communities engaged in relocalized and regenerative production are not in sync with the mutual coordination capacities developed in the crypto/web3 context.

The idea here is a potential ‘entanglement’ between the local and the translocal level, which creates new levels of strength and capacity for the local.

Otto et al. (2019) point out that instead of focusing on the poorest, efforts aimed at reducing carbon emissions should target people at the top of the social scale, i.e., the wealthiest.

I feel like sharing with you some of my observations as a frame analyst, as someone who analyzes semantic frames and how they structure a discourse to, in this case, to disempower us and keep us embedded within a conversation that is primarily about the actions of corporations and nation-states and that disengages us from direct grassroots action and taking power into our own hands

we're using post in the way postmodernists use post, which is it's informed by modernism, it's informed by capitalism without being able to transcend it necessarily because capitalism and it's the most recent incarnation of capitalism, which is neoliberalism, is like the oxygen that we breathe. It's all encompassing. It's totalitarian in its nature. And it's pervasive. And so in that sense, we say we have to be informed by the logic of the dominant system.

The potential political power of young people has been demonstrated

By thenscientists need to have improved their communication with the public, especially with youngpeople

you're synchronizing with that time

Romanticizing any society as purely altruistic undermines their sophistication and resilience.

Fortunately, we do not have to agree on everything. Linked Data enables layered agreements, in which a few rules need to be adopted by many, and sets of additional rules are agreed upon by smaller groups as required.

Freedom of course always comes at a cost: what constitutes a victory for personal rights and freedom of speech also facilitates the spread of illegal messages, since decentralized networks make it harder to control what information is exchanged. Legality is of course a tricky matter, as some countries instate laws that prevent their citizens from voicing opinions that would be legal elsewhere.

Meeting Purpose Explore the potential of Fair Shares Commons and strategize on implementation in Kansas/Missouri region.

I’m

she says again, her voice now tinny

Aunt Lydia stands up, smooths down her skirt with both hands, and stepsforward to the mike. "Good afternoon, ladies," she says, and there is aninstant and earsplitting feedback whine from the PA system. From among us,incredibly, there is laughter.

opology has reached the point where a mathematician engagedin topological research is not only justified in calling himself a topologist,but he must specify whether he is a point set topologist, differentialtopologist, algebraic topologist, or some other topological specialist.

Rufus-scheduler (out of the box) is an in-process, in-memory scheduler. It uses threads. It does not persist your schedules. When the process is gone and the scheduler instance with it, the schedules are gone.

We focus on cities and businesses because of the magnitude of their impacts on the Earth system, and their potential to take swift action and act as agents of change.

“Accuracy” indicates the closeness of a measurement to the true value but it cannot be quantified because it is not possible to know the true value.

Sure, it is not needed, we can always write things in a different way. As a matter of fact, with such an argument, hardly any improvement should be accepted.

often I get the question, what should we do? And they expect  me to talk about um, mobility and, um how to reduce flying and  all forms of consumer choices. And they get surprised when I say  that the number one issue is talk to your friends.

if you could get everyone on the planet to do one thing what would it be and she said stay exactly where you are and figure out 00:33:30 what it is that you can do in your local

I’ll pull out a few key points as to why I think this approach is far too simplistic to be meaningful

Who says it's not a word? Not a word, simply because lexicographers have not recognized it? When a lexicographer recognizes it, it has already been in use! Even Mr. Fiske says it is a word, although he obviously disprefers it.

I believe it is possible to disprefer something while either 1. not disliking it, or 2. liking it but not intensely enough to be the preference. As in, "I like tart apples, but I sometimes disprefer them as an ingredient on a green salad." It doesn't and hasn't, meant I would refuse to eat a salad with this ingredient included, but there are times when my preference would have been to have a salad without them.

I think you linguists worry too much. It's a simple enough formation using a very common prefix, and while it is not clear whether "I disprefer" means "I do not prefer" or "I prefer something other than" or "I prefer the opposite of" or "I stop preferring", either it'll settle down to one meaning or it'll carry a range. So what? This is the first time I've heard the word but I don't find it particularly puzzling.

If you said anti-prefer, I'd have a better idea of what the word meant.

Amy: It's a real word. I use it all the time (of course, I'm a linguist, and I allow the possibility that I picked it up from my linguist chums, though it doesn't seem particularly jargony to me). For me, "disprefer X" means something like "not choose X when other options are available". This is subtly different from "prefer anything over X", quite different from "not prefer X", and totally distinct from "dislike X" or "object to X".

the model Waits are just a large files of numbers on a server and these can be easily stolen all it takes is an adversary to match your trillions 00:41:14 of dollars and your smartest minds of Decades of work just to steal this file

en réalité l'efficacité ce qui ce qui me paraît surtout intéressant à voir c'est que l'efficacité même du redoublement ne signifie pas la même 00:34:56 chose pour les chercheurs et pour les enseign plutôt que de se demander qui a raison et bien sûr les chercheurs ont leur logique propre et encore une fois il s'agit de de faire un raisonnement quasi expérimental on peut comprendre mais ça 00:35:09 veut dire que la question des finalités du redoublement est posée différemment par les enseignants et par les chercheurs et et en fait ce qu'il faut voir c'est que la manière dont les chercheurs évaluent l'efficacité d'un redoublement est beaucoup plus exigeante 00:35:22 que ce qu'on attendent les enseignants pourquoi parce que ce qu'on attend enseignant du redoublement c'est pas qu'un élève parvienne à à performer de sorte à aniler ces difficultés ù à ne 00:35:36 plus avoir de difficultés à long terme c'est surtout de le remettre en selle et donc c'est de chercher un effet de de boost un effet à court terme et donc cet effet c'est typiquement des choses qu'on constate dans les études cet effet il 00:35:49 est il est là mais les chercheurs vont dire bah oui mais l'effet se maintient pas à long terme bon et donc en quelque sorte ils attendent du redoublement quelque chose que les acteurs de terrain n'en attendent pas donc ils sont plus 00:36:00 exigeants ce qui veut dire qu'au fond la manière même de mesurer l'efficacité elle est pas vraiment elle est pas vraiment neutre en quelque sorte elle elle est porteuse d'une certaine conception d'une certaine valeur et ben 00:36:14 une autre manière de de percevoir à quel point la la méthodologie est porteuse d'un certain nombre de de conceptions de la de la finalité c'est que au fond dans 00:36:27 les études seul le redoublement est considéré comme le traitement hein donc on est dans une dans un paradigme où au fond on on regarde ce qui se passe lorsque les certains élèves se voient appliquer le traitement du redoublement 00:36:39 et implicitement on considère que la promotion automatique ne serait pas un traitement la promotion automatique n'aurait pas à être considérée comme un traitement c'est comme si 00:36:50 euh au fond euh il était enfin voilà en fait euh si on réfléchit dans un système où euh bah l'avancement dans la scolarité est conditionné par l'évaluation par l'acquisition des 00:37:05 apprentissage et la et le fait de démontrer qu'on a acquis les apprentissages ben imposer du jour au lendemain une promotion automatique ça n'a rien de neutre ça n'a rien de neutre du tout c'est quand même euh quelque chose qui pourrait même paraître assez 00:37:16 aberrent donc ici euh au fond dans dans les études on se pose pas du tout euh ce genre de question

c'est pas juste un problème de de conception enfin de réception parce que on se dit bah la réforme a été bien 00:48:16 conçue le problème c'est une fois qu'on essaie de l'appliquer non le problème c'est qu'elle a été conçu aussi d'une manière qui me semble-t-il est problématique et donc la conception est bien en cause et ça ça plaide aussi pour 00:48:28 mieux intervenir enfin mieux comprendre les contextes avant d'intervenir sur les contextes pour éviter justement que ces contextes en quelque sorte se retournent contre la politique et c'est un enjeu 00:48:39 d'efficacité mais c'est aussi un enjeu me semble-t-il démocratique à l'air aujourd'hui où les preuves chiffrées sont de plus en plus preignantes et de plus en plus valorisées mais aussi où elles sont de plus en plus fragiles au 00:48:53 sens où il y a une forme de scepticisme croissant dans l'opinion à de l'objectivité quantitative et donc voilà c'est des choses qu'on peut pas totalement ignorer me semble-t-il si on veut en tout cas être dans une 00:49:05 perspective pragmatique de se dire comment est-ce qu'on peut améliorer au fond les le système à l'aide des connaissances produ produites par la

la littérature elle pointe certaines limites des dispositifs participatifs quand ils sont initiés par d'autres acteurs que la population elle-même et notamment par les pouvoirs publics euh avec par exemple des auteurs qui montrent que dans dans certaines politiques la la poli la participation peut devenir une injonction avec des formes de participation à libi donc c'està-dire avec une certaine instrumentalisation de cette participation en vue de légitimer l'action publique donc ça c'est c'est un peu des des travaux critiques et qui qui montrent des logiques plutôt descendantes

il existe aussi des travaux qui montrent que ce sont souvent les les catégories sociales les plus favorisés qui ont le plus facilement accès au dispositifs et aux instances participatives donc finalement des dispositifs qui vont un petit peu reproduire des inégalités sociales alors que la participation est censée les réduire

what's the point what am i g to get out of this it's the same question actually

l'idée n'est pas évidemment dans le cadre du 00:03:23 développement des compétences psychosociales d'aller jusqu'à un lissage une neutralisation de des comportements mais plutôt de permettre que ces expressions d'émotion se 00:03:37 produisent dans un dans un cadre qui permette justement que ces émotions soit entendue et puissent être prises en compte dans le milieu

si on développe des compétences chez les enfants les adolescents ou les adultes des compétences a exprimé des besoins à porter des revendications à aller interpeller des niveaux de décision et 00:01:51 bien encore faut-il que le système permettent à la fois l'expression de ses compétences mais aussi que lorsque ces revendications sont portés à un niveau décisionnel et bien on leur claque pas la porte au nez

développer les compétences psychosociales des enfants ou des adolescents en milieu scolaire pose la question de la prise en compte du 00:01:13 système scolaire dans lequel il évolue au sens large est par exemple on peut bien imaginer que développer les compétences psychosociales des enfants et des adolescents dans un établissement scolaire où règne un climat scolaire 00:01:27 épouvantable n'a pas beaucoup de sens

l'intervention de lecture parentale que je vous ai présenté s revéé inefficace pour les familles àopon c'est la suggè 01:01:30 peut-être que il est important de travailler à des interventions qui lève les barrières linguistiques pour les parents concernés

ces interventions doivent être évidemment complété par d'autres types d'action de nature plus structurelle par exemple dans le cas des inégalités d'accès aux crèches que je viens de mentionner il est nécessaire aussi 01:01:17 d'intervenir sur les critères d'éligibilité sur la répartition territoriale de l'offre

lorsqu'on retrouve des impacts positifs il se concentre principalement sur des on retrouve des 00:37:11 impacts positifs principalement lorsque ces études ne ciblent pas des familles socialement défavorisées par contre quand on cible des familles socialement défavoriser les impacts sont bien moins 00:37:25 importants

The cell phone providers usually call them "mobile" phones which is more precise since "cell" refers to a kind of technology.

However, it is increasingly becoming just a "phone", as landlines continue to disappear from households.

I wouldn't focus too much on "posted only after human review" - it's worth noting that's that's worth nothing. We literally just saw a case of obviously riduculous AI images in a scientific paper breezing through peer review with noone caring, so quality will necessarily go down because Brandolini's law combined with AI is the death sentence for communities like SE and I doubt they'll employ people to review content from the money they'll make

I got no actual help from my long Verizon Support chat session and I kept asking if there is a block list they use that they could check (or a whitelist I could be added to...but fat chance) my IP for, since that is clearly what the error is calling out, but they never acknowledged that particular part of my questions, just ignored it.

The Turning Point’

si je peux me permettre juste de revenir sur cette question de 00:08:58 rajeunissement ce qui est sûr c'est que il est plus visible ce rajeunissement puisque avant on ne faisait pas d'enquête dans l'école primaire donc ces chiffres sont tout nouveau en fait donc ça donne l'impression qu'il y a énormément de violence

To prevent accidental unsubscriptions, senders return landing pages with a confirmation step to finish the unsubscribe request. A live user would recognize and act on this confirmation step, but an automated system would not. That makes the unsubscription process more complex than a single click.

Ce passage à haut risque nécessite donc une attention et un accompagnement spécifique pour les élèves de milieu populaire et en particulier pour ceux issus de l’éducation prioritaire.

doit différencier résultats, travail et attitude

le jugement ne doit en aucun cas porter sur la personne

le conseil doit proposer des pistes à l’élève pour progresser

ensemble du bulletin est rouge et orange

Par ailleurs, elle met au même niveau des compétences qui peuvent être variées et non compensables

Lorsque la note est absente, si les bulletins ne présentent que des commentaires, les échanges portent alors sur l’interprétation de ces commentaires. Cela oblige les enseignants à être plus explicites. En revanche, cette lecture prend du temps et l’interprétation des commentaires dépend largement de l’acculturation du lecteur au milieu scolaire.

il faut penser à la clause de confidentialité, cela suppose de rappeler ce principe et de demander un engagement des participants

sous réserve qui est des règles de gouvernance qui permettent d'éviter le phénomène de passagers clandestin

I find it ridiculous that we spend energy on debating whether an alternate spelling is "correct" - real people, not English professors and dictionary authorities, are the authorities on English-as-used, and will ultimately make the distinction irrelevant.

nowadays many people work with docker containers. Most default docker images do not have bash and something like [[ $string == *foo* ]] will not work.