La burbuja de la IA

[Cadavre Exquis]

https://x.com/karpathy/status/2056753169888334312

OpenAI Co-Founder Andrej Karpathy Joins Anthropic
Posted by BeauHD on Tuesday May 19, 2026 @04:00PM from the AI-hiring-wars dept.

OpenAI co-founder Andrej Karpathy has joined rival AI lab Anthropic. "The hire is a major coup for Anthropic in the high-stakes competition for elite AI talent -- and another sign the company is emerging as a magnet for some of the industry's most respected technical minds," reports Axios. From the report:
Karpathy will start this week on Anthropic's pre-training team, which is responsible for the massive training runs that give Claude its core knowledge and capabilities, according to Anthropic.

Karpathy will help launch a new team focused on using Claude itself to accelerate pretraining research -- an increasingly important frontier as AI companies race to automate parts of AI development. "I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D," Karpathy said in a post on X.

Karpathy is a rare AI figure with credibility across research, industry and education. He was a founding member of OpenAI before serving as Tesla's director of AI, where he led the computer vision team behind Autopilot. Karpathy coined the term "vibe coding" and recently described himself as being in a "state of AI psychosis" since December -- embracing "tokenmaxxing" and aggressively stress-testing frontier models.

Saludos.

[Benzino Napaloni]

Un estudio lo confirma: la inmensa mayoría de empresas que han despedido empleados y los han reemplazado por la IA no han mejorado sus beneficios https://share.google/AyFZ1RmyrQUAJANR7

Ya hay un nombre para esto: AI Washing.

https://www.linkedin.com/posts/aeejazkhan_a-software-engineer-at-atlassian-was-laid-share-7461016199250919424-zXj_/

A software engineer at Atlassian was laid off in March after 8 years with the company.

His response was a detailed 38-minute YouTube video that reveals how the company’s entire web traffic architecture works, now available for free for anyone to learn from or copy.

His name is Vasilios Syrakis. He worked in Sydney on Atlassian’s “digital plumbing” — the critical system handling web traffic with about 2,000 programs running across 13 regions worldwide.

Every time a user clicks into Atlassian’s products, Syrakis’s system decides which servers respond. Atlassian’s own engineering blog had featured his team’s work as recently as February 2025.

That same quarter, Atlassian reported record revenue of $1.79 billion. Its cloud business grew 29% year over year, with 350,000 customers including 80% of the Fortune 500.

Despite these strong numbers, the company laid off about 10% of its staff. Leadership described the cuts as necessary to “self-fund AI investment.”

In the six months before the layoffs, CEO Mike Cannon-Brookes sold 866,145 shares for roughly $134 million. Co-founder Scott Farquhar sold the exact same number on the same schedule.

The board also approved $2.5 billion in stock buybacks to support the share price. Even so, Atlassian shares have fallen 56% this year.

This pattern has fueled skepticism around “AI washing.” While companies cite AI as a reason for cuts, many layoffs stem from other financial priorities.

Sam Altman recently highlighted this issue. Of the 1.2 million American jobs cut in 2025, only 55,000 were directly attributed to AI.

Now without a paycheck to protect, the engineer who helped build Atlassian’s core infrastructure is openly teaching the world how it works, for free.

https://lesluthiers.org/verversion.php?ID=315

MM: “... y miles de personas pasarán hambre, y algunos gobernantes se enriquecerán a costa de su pueblo”
JM: Perdón, ¿una guerra cada siglo?, ¿nada más?
DR: ¿Miles de personas pasarán hambre?, ¿nada más que miles?
LP: ¿Sólo algunos gobernantes?

TODOS: Quédate niño,
quédate acá...

 


Es decir, de 1,2 millones de despedidos en EEUU el año pasado, sólo 55.000 han sido realmente despedidos por amortización por IA. De las empresas que dicen que despiden por IA el relato es falso (despiden porque hay que pagar la IA, que vale un cojón), o falsísimo (empresas que se están hundiendo pero para disimular dicen que despiden para adoptar la IA).

Esta burbuja va a dejar a la de las puntocom a la altura de una broma.

[muyuu]

https://blog.google/products/ads-commerce/google-marketing-live-search-ads/

Todo ese ladrillo que acaba de publicar Google es para venderte que te empiezan a poner anuncios en las respuestas de la IA. Por supuesto te sacan toda la información posible y comercian con ella, lo admitan o no, y ya pueden investigar y extraer tus archivos a escala como nunca antes.

[Saturio]

La presentación de Google 1/0 fue tremendamente deprimente.
En el vídeo de Youtube han tenido que bloquear comentarios. Hay otro vídeo resumen que tiene los comentarios abiertos (o los tení ayer por la noche).

Es un chorrear de nuevos servicios y servicios sobre servicios todos integradísimos en servicios tutelados por la IA.
Un montón de cosas que nadie ha pedido pero que parece que van a desplazar a las que tenemos y que con sus defectillos y su, ya rampante, enshittificación más o menos funcionaban.
Todo recomendaciones realizadas a través de los datos que ya tienen y los que queriendo o no vas a entregar a la IA.
Como decía algún comentario, enseñaremos a los nietos las fotos de juventud que la IA generará por nosotros (sin que se lo pidamos) sobre experiencias que no hemos tenido con gente con la que nunca hemos estado.
Parece que quieren que nos levantemos de la cama con una agenda repleta de cosas por hacer y comprar que ni siquiera nosotros hemos decidido porque la IA sabe más que nosotros lo que nos conviene.

Lo dicho, deprimente.

[muyuu]

Las compañías que recortan empleos para sustituirlos por IA perderán contra las compañías que no lo están haciendo. : https://libertas.software/en/knowledge-hub/19/the-companies-cutting-headcount-for-ai-will-lose-to-the-ones-who-didnt

There is a version of AI adoption that looks smart on a spreadsheet. Fewer people, lower payroll, same output. It is the version being quietly executed in boardrooms right now, dressed up in language about efficiency and transformation.

It is also the version that will cost those organisations dearly over the next five years.

This is not an argument against AI. It is an argument for using it correctly — and the distinction matters more than most leadership teams currently appreciate.

The Asset They Are Cutting Is the One They Cannot Rebuild

When an organisation downsizes in response to AI capability, the assumption is that the work being removed was the value. That the task itself — the report, the analysis, the email, the data entry — was what the role existed to do.

That assumption is wrong.

The real value sitting inside most teams is not the work they produce. It is the knowledge they carry. How the business actually operates. Where the edge cases live. Why certain decisions get made the way they do. What customers really mean when they complain about a specific issue. The context that never makes it into a process document because it does not need to — because the right person already knows.

That knowledge is institutional. It is built over time. It is extraordinarily difficult to reconstruct once it walks out the door. And right now, organisations are letting it go in exchange for short-term cost reductions, without fully accounting for what they are losing.

AI Does Not Replace Judgement. It Multiplies It.

The organisations that will come out ahead are not the ones who used AI to do the same work with fewer people. They are the ones who used AI to do significantly more work with the same people — or with people who are better positioned to apply their judgement at scale.

This is a fundamentally different operating model. Instead of replacing a team member's output, AI extends their reach. A marketing team that previously managed one campaign at a time can now manage five. An analyst who spent three days on a report can now produce one in a morning and spend the rest of the week on interpretation and strategy. A customer success manager who handled thirty accounts can now meaningfully engage with a hundred.

The human is not removed from the equation. The human is the equation. AI is what makes that equation run faster.

Business Knowledge Is a Competitive Advantage — But Only If You Keep It

There is a compounding effect to institutional knowledge that does not show up in headcount metrics. Experienced teams make better decisions. They catch problems earlier. They understand the business deeply enough to apply new tools — including AI tools — in ways that actually fit the organisation's context.

An AI system is only as useful as the judgement that guides it. A prompt written by someone who deeply understands the customer base, the product, and the operational constraints will produce something categorically more valuable than the same prompt written by a replacement hire working from a brief. Context is not a soft advantage. It is a hard one.

When organisations cut experienced team members in favour of AI-led efficiency, they often discover too late that the AI works considerably better when the people who truly understand the business are the ones directing it.

The Right Question to Be Asking

Rather than asking "where can AI replace people?" the more useful question is: "where can AI give our people back the time they are losing to tasks that do not require their judgement?"

Most organisations have a significant amount of high-skill time absorbed by low-skill work. Administration, formatting, scheduling, basic reporting, first-draft production. These are areas where AI can deliver genuine relief — not by removing roles, but by removing the friction that stops experienced people from operating at their best.

The teams that reclaim that time and redirect it toward the work only they can do — relationship management, strategic thinking, complex problem solving, nuanced decision making — will have a meaningful edge. Not because they have fewer costs. Because they have more capability.

A Sustainable Model Looks Different

Done well, AI adoption should result in teams that are more effective, more focused, and more capable of delivering at a level that was not previously achievable. It should make the knowledge inside an organisation more accessible, not more redundant.

The organisations that understand this will invest in training their teams to work alongside AI tools rather than replacing teams with them. They will treat business knowledge as infrastructure. They will build processes where AI handles the volume and humans handle the depth.

That is not a more cautious version of AI adoption. It is a more ambitious one. Because it is asking AI to do something harder than replacing human output — it is asking it to multiply human potential.

The companies currently cutting headcount to absorb AI costs are making a short-term trade with long-term consequences. The ones holding their teams together and investing in how those teams operate with AI are building something more durable.

The gap between those two approaches will become visible sooner than most expect.

[bluebeetle2]

Linkedin (Microsoft) ha cerrado la sede de Linkedin Learning en Graz en su proceso de optimización y ha despedido a un 5% de la plantilla. La causa de cara a la galerIA es la de siempre. La realidad es un crecimiento excesivo basado en expectativas irreales. Había una fuerte presencia española ahí.

https://www.businessinsider.com/linkedins-layoffs-come-with-operational-changes-memo-shows-2026-5
https://www.reuters.com/business/world-at-work/linkedin-is-planning-lay-off-5-staff-latest-tech-sector-cuts-source-says-2026-05-13/

La IA es el bálsamo de fierabrás que todo lo cura, todo lo explica, para todo vale y para nada sirve.

[muyuu]

https://archive.ph/WfCta ( https://www.theverge.com/tech/930447/microsoft-claude-code-discontinued-notepad )

Microsoft starts canceling Claude Code licenses


Microsoft first started opening up access to Claude Code in December, inviting thousands of its own developers to use Anthropic’s AI coding tool daily. It was part of an effort to get project managers, designers, and other employees to experiment with coding for the first time, and sources tell me that Claude Code has proved very popular inside Microsoft over the past six months. Perhaps a little too popular, as Microsoft is now preparing to walk back its Claude Code push.

I understand that Microsoft is planning to remove most of its Claude Code licenses and push many of its developers to use Copilot CLI instead. While Claude Code has been a popular addition, it has also undermined Microsoft’s new GitHub Copilot CLI coding tool — a command line version of GitHub Copilot that runs outside of development apps like Visual Studio Code.

I’m told that Microsoft’s Experiences + Devices team, which includes the engineers responsible for Windows, Microsoft 365, Outlook, Microsoft Teams, and Surface, is winding down its usage of Claude Code by the end of June. Sources tell me that engineers are being encouraged to start transitioning their workflows to GitHub Copilot CLI in the coming weeks, ahead of the cutoff.

Microsoft is telling employees that the decision is about converging on Copilot CLI as its main agentic command line interface tool across Experiences + Devices, but sources tell me the decision is also a financial one. The June 30th cutoff is the last day of Microsoft’s current financial year, and canceling Claude Code licenses is an easy way to cut some operating expenses for when the new financial year starts in July.
“When we began offering both Copilot CLI and Claude Code, our goal was to learn quickly, benchmark the tools in real engineering workflows, and understand what best supported our teams,” says Rajesh Jha, executive vice president of Microsoft’s experiences and devices group, in an internal memo seen by Notepad. “Claude Code was an important part of that learning… at the same time, Copilot CLI has given us something especially important: a product we can help shape directly with GitHub for Microsoft’s repos, workflows, security expectations, and engineering needs.”

The transition away from Claude Code won’t be an easy one for engineers inside Microsoft, though. Microsoft had been encouraging employees without any coding experience to experiment with Claude Code, allowing designers and project managers to prototype ideas. Microsoft had also originally expected employees to use both Claude Code and GitHub Copilot, to compare the two and provide feedback.

Microsoft’s own developers have favored Claude Code over GitHub Copilot CLI in recent months instead, and there are still gaps between the products that will now need to be addressed. Microsoft had reportedly considered acquiring Cursor in recent months to help close the GitHub Copilot gap, but has started looking at different AI startups to bolster its AI ambitions and avoid potential regulatory scrutiny.
“We are partnering closely with GitHub and continue to improve Copilot CLI for Microsoft engineers,” says Jha. “The GitHub team has already shipped significant improvements based on Microsoft feedback, and Experiences + Devices will remain closely involved in shaping the product. This is a shared accountability across GitHub and E+D leadership: to make Copilot CLI the best agentic coding experience for Microsoft engineers.”

Anthropic’s models will remain accessible through Copilot CLI, along with internal-only Microsoft models and OpenAI’s range of models. I understand that Microsoft is planning to invest more in Copilot CLI so it’s deeply integrated into Microsoft’s own engineering workflows. Microsoft is also encouraging developers to file bug reports and feedback on Copilot CLI ahead of Claude Code being removed.
Microsoft quickly became one of Anthropic’s top customers earlier this year and has even reportedly been counting selling Anthropic AI models toward its own Azure sales quotas. Microsoft also signed a deal with Anthropic in November that allows Microsoft Foundry customers to get access to Claude Sonnet 4.5, Claude Opus 4.1, and Claude Haiku 4.5.

The decision to cancel Claude Code licenses won’t have any impact on the Foundry deal, and Microsoft still continues to favor Anthropic’s Claude models inside Microsoft 365 apps and Copilot, where they’re more capable at certain tasks than OpenAI’s counterparts. Microsoft also worked closely with Anthropic recently to bring the technology behind Claude Cowork into Microsoft 365 Copilot.
The pressure is now on Microsoft’s GitHub team to improve Copilot CLI and try to surpass Claude Code in the process. Microsoft told me last year that 91 percent of its engineering teams were using GitHub Copilot, but Claude Code usage over the past six months has definitely had an impact on that number. Microsoft now wants to turn GitHub Copilot usage around and have its own engineers once again improving its own AI coding tool.