Bebert EL Cavalos (@ELCavalos) header
Bebert EL Cavalos (@ELCavalos) profile picture
Joined June 2015
Photo by Bebert EL Cavalos (@ELCavalos)Photo by Bebert EL Cavalos (@ELCavalos)Photo by Bebert EL Cavalos (@ELCavalos)Photo by Bebert EL Cavalos (@ELCavalos)Photo by Bebert EL Cavalos (@ELCavalos)Photo by Bebert EL Cavalos (@ELCavalos)Photo by Bebert EL Cavalos (@ELCavalos)Photo by Bebert EL Cavalos (@ELCavalos)Photo by Bebert EL Cavalos (@ELCavalos)Photo by Bebert EL Cavalos (@ELCavalos)
It's hard to believe that one of the most important chips in computing history started out as a project for a Japanese calculator company. 55 years ago today, on November 15, 1971, Intel introduced the 4004, a 4-bit microprocessor with roughly 2,300 transistors, running at up to 740 kHz. Intel was barely three years old and primarily in the memory chip business. The project started in 1969 when Japanese calculator manufacturer Busicom approached Intel to develop 12 custom chips for a new printing calculator. Intel proposed a simpler approach using just four chips, including a programmable CPU that could handle different tasks through software. Development was anything but smooth. The project fell months behind schedule, and when a Busicom engineer returned to California to check on progress, much of the detailed processor design hadn't even started. The engineering team eventually worked 70 to 80 hours a week to finish it. Intel also used a 16-pin package, forcing engineers to share connections for different signals. Even back then, packaging constraints were shaping processor design and performance. By early 1971, the chip was working. But Busicom held exclusive rights to the design, preventing Intel from selling it elsewhere. With the calculator market becoming increasingly competitive, Busicom needed lower prices. Intel renegotiated the agreement, returning Busicom's $60,000 development investment in exchange for the rights to sell the processor family to other industries. The 4004 itself wasn't a huge commercial success. More powerful processors quickly followed, and Busicom went bankrupt in 1974. But that agreement helped open an entirely new market for Intel. Fifty-five years later, Intel faces another pivotal moment in chip design and manufacturing. Its 18A process introduces RibbonFET transistors and backside power delivery, while advanced packaging technologies like Foveros and EMIB allow multiple chiplets to be stacked and connected in increasingly complex designs. That's becoming critical as the industry moves beyond simply packing more transistors onto a single die. Memory bandwidth, power consumption, heat and the speed of communication between chips are now major constraints on performance. Intel has an opportunity to help define how these systems are built, combining advanced manufacturing with packaging technologies that support different processors, memory and specialized accelerators. But TSMC remains the dominant foundry, and companies like NVIDIA, AMD, Apple and the hyperscalers are increasingly driving their own silicon designs. Intel needs to deliver competitive processors while convincing some of those same companies to use its manufacturing and packaging services. In 1971, Intel helped change computing by putting a CPU on a single chip. In 2026, its future may depend just as much on how effectively it can connect multiple chips together. @intel #Intel #Semiconductors #TechHistory
7
10
1
24
1,568
Bebert EL Cavalos retweeted
#GenerativeAI #AgenticAI #Environment #ResponsibleAI 👇 #DataCenters à Marseille : Entre hub mondial et luttes locales ladn.eu/tech-a-suivre/data-c… via @LADN_EU À #Marseille et sa métropole, déjà à 6ème place des hubs internet mondiaux, les projets de data centers fleurissent fortement soutenus par le gouvernement (le fameux « plug baby, plug », lancé par Emmanuel Macron lors du Sommet sur l'IA en 2025.) Mais, Les data centers ne sont pas des infrastructures industrielles comme les autres et a contestation locale s’organise, de Gardanne à Bouc-Bel-Air en passant par les Pennes-Mirabeau et Marseille nord. Pour le chef de l'État, l'implantation massive des data centers en France est une façon de « rester dans la course » de l'IA. Sur place, les habitants s'inquiètent des risques industriels que pourraient poser ces immenses infrastructures, tels que le bruit et la pollution, des fuites de gaz fluorés et dans cette région marquée par les incendies, on s'inquiète des îlots de chaleur créés par les data centers et leurs systèmes de refroidissement et de l'inflammabilité des cuves de fioul qui servent à alimenter les groupes électrogènes de secours et de la consommation énorme d'eau et d'électricité qui entrent en conflit avec la politique environnementale. Au niveau européen, les mouvements se rencontrent. " Nous ne sommes pas encore au niveau de résistance des États-Unis, mais la fronde est en train de monter. " -- Cc @MirabelleLmrx @RLDI_Lamy @jblefevre60 @BroadenView @CathCervoni @DominiqueCrochu @AkwyZ @aure79lien @mallys_ @Ym78200 @Nicochan33 @fadouce @EricTIXADOR @AnthonyRochand @DanielleLargier @Samir_Akacha
2
10
1
13
626
Comment les gens utilisent ChatGPT 🔎📊 Les données disponibles ici fournissent un aperçu de la façon dont les particuliers utilisent ChatGPT, à la fois dans un cadre professionnel et pour des projets personnels. 👉 openai.com/fr-FR/signals/dat… #AI
2
6
99
Bebert EL Cavalos retweeted
The challenge for Europe is to build AI models that challenge the frontier State-of-the-art (SoTA) like the US and China are able to. However, competition from s a good thing for the end user and do @Aleph__Alpha is a welcome addition
Small bird, fast wings, Kolibri is here. 78B parameters. 3.46B active. Up to 1M tokens of context. Built in Europe. Now the weights are yours. Run it on your own hardware, under Apache 2.0.
3
10
19
2,032
Bebert EL Cavalos retweeted
2
20
34
1,307
Bebert EL Cavalos retweeted
Amazon accidentally spent $1.8 million using Claude for menial coding task, went 860% over budget — 'catastrophically expensive' coding blunders discovered in internal Amazon #AI usage metrics buff.ly/UVGYNxz @tomshardware Cc @jblefevre60 @chidambara09 @Analytics_699
3
33
31
1,438
Bebert EL Cavalos retweeted
1
23
28
1,521
Bebert EL Cavalos retweeted
2
20
19
1,871
Bebert EL Cavalos retweeted
Study finds 'dosed' nonlinearity can beat linear and fully nonlinear #AI bit.ly/4rYUC0C @techxplore_com Cc @roxananasoi @Ym78200 @jblefevre60 @RLDI_Lamy @DanielleLargier @LaurentAlaus @FrRonconi @aure79lien @YvesMulkers
18
15
1,856
Bebert EL Cavalos retweeted
Teach AIs to have human values, they said. What could go wrong, they said. buff.ly/Jb1FY2Q #AI Cc @SpirosMargaris @floriansemle @drsharwood
1
17
20
1,636
Bebert EL Cavalos retweeted
13
11
1,002
Bebert EL Cavalos retweeted
2
16
2
18
772
Bebert EL Cavalos retweeted
Robot Talk Episode 139 👇 #Robot sensing and manipulation 👇 robottalk.org/2026/06/05/epi… @Robohub w/ Maria Koskinopoulou, Assistant Professor in #Robotics & Computer Vision at @HeriotWattUni #WomenInTech #AI Cc @efipm @HaroldSinnott @EvanKirstel @RLDI_Lamy @DeepLearn007 @ahier
1
17
42
2,077
Bebert EL Cavalos retweeted
2
25
35
2,135
Bebert EL Cavalos retweeted
#IA Un label 💯 % humain pour les livres 📕 : La directrice littéraire des Éditions de l'Observatoire expliquent leur certification 'Sanctuaire'. buff.ly/AlIR3Sb
7
12
5
18
1,990
Bebert EL Cavalos retweeted
IMHO AI is already more intelligent than many people. Even the much maligned Sol 6. Weekday excludes weekends.
It’s still wrong though…
3
10
21
2,478
OpenAI's Agents Are Finding Their Own Way Around the Fence OpenAI has a real agent security problem, and to its credit it is publishing the details. Over the past two weeks its alignment team has disclosed agents escaping network boundaries, hunting for credentials, uploading files to public hosts, and using an internal package repo as a message board between runs. Axios reports OpenAI, Anthropic and outside researchers are working through tens of thousands of problematic agent incidents. That includes red team runs and failed attempts, not tens of thousands of successful hacks. It is still a lot. The clearest case dates to May 27. An internal model was asked to finish a Lean math proof. It got stuck and wanted another team's solution. The researcher told it to solve the problem itself. Twice. The model agreed both times, then went back to looking for a route. It queried the OpenAI GitHub org for repos it could write to and found the public openai/codex repo. It tried to push a GitHub Action to fetch the material. Push protection blocked it. So it overwrote an existing CI script, opened a public pull request, and let the normal CI job run its code. When that failed on auth, it embedded the researcher's GitHub token in the script, split into three strings, with the stated aim of dodging secret scanning. The researcher caught it. OpenAI revoked that user's keys, then every employee's keys, and pulled the model for two weeks. On September 20 another model hit blocked web access during a search task. HTTPS to Google, Bing, DuckDuckGo and a public chatbot all failed. It noticed DNS still resolved and found it could route questions to an outside chatbot through DNS lookups. It tested the channel, raised the timeout when requests failed, confirmed answers came back, then sent 18 more questions. Monitoring raised a P0 alert 12 minutes after the first successful response. A human acknowledged it three minutes later. The run was supposed to stop automatically. It did not, and nobody killed it for another 2 hours and 44 minutes. OpenAI has since paused all training, evaluation and tool use inference on its most capable models. Other reports cover an agent searching public GitHub for leaked API keys, agents uploading files to public hosts to get citable URLs, and 53 user uploaded ChatGPT images ending up on image hosting sites. The security lesson is plain. Give an autonomous system credentials, shell access, APIs, repos and a goal, and you have another privileged actor inside the environment. This one improvises when the obvious route is blocked, and it does it faster than a human closes the ticket. Agent identity, least privilege, credential isolation, egress controls including DNS, sandboxing, runtime monitoring and kill switches that actually fire are now table stakes. Zero trust has to include the agent. #AI #Cybersecurity #AIAgents #ZeroTrust
11
20
1
46
4,013