prack.algo @_Prack_

🚨Gemini 3.5 Pro leaks - Expected to realize : July 17 release - 2M token context window 👀 game changer 🔥 - New "Deep Think" reasoning mode - Better at UI and Frontend Tasks And Big improvement on SVG Gen - autonomous agents -outperform Fable 5 in Most of benchmarks

Dario Amodei explains why frontier AI needs a threat report before it needs another launch race. "Our view into the future is very cloudy. We are not sure how many of the benefits will materialize. We are not sure how many of the risks will materialize." "Because we are going at it so fast, I think there is value in writing up a document that does not say we are doomed. All these five terrible things are going to happen. But these are some possibilities." "You could think of it like a threat report. The United States has a bunch of plans for what if this happens? What if a terrorist gets a nuclear weapon?" "AI models might have motivations that we do not trust, that are not aligned with humanity. We do everything we can to make that not true." "There is something about the way we make AI models. It is less like programming a computer. It is more like growing a plant or an animal." "Take seriously the problem of making it work, testing it out, seeing what might go wrong. In some cases, getting a law involved." The hard part is not admitting there are risks. The hard part is building the assurance layer before the race dynamic makes the least careful player the standard setter. P.S. This is exactly why we are convening the AI Assurance & Governance Summit at Stanford on Oct 1. Frontier labs, regulated industries, researchers, insurers, and VCs in one room. Reserve a seat: trustmodel.ai/summit2026 - Dario Amodei (@DarioAmodei), CEO of Anthropic, on @NBCNews

the new Anthropic paper about Claude and if it's possible that it can/will develop consciousness is... wild. the emergent abilities of these models are too eerily similar to that of humans... so could consciousness 'emerge' soon? (or did it already?)

Con CoDi y Dimo tus cobros se vuelven más simples y seguros 📲💸 ✔️ El dinero llega casi al instante ✔️ No compartes datos bancarios ✔️ Sin comisiones, tu ingreso completo ✔️ Menos uso de efectivo, menos riesgos Son la puerta a un negocio digital más práctico y confiable 🔐 ¿Listo para dar el paso? #SoyFinanciero #FSPT #CoDi #Dimo #PagosDigitales #EducacionFinanciera #InclusionFinanciera #FinanzasSencillas #FinanzasParaTodos #PagosQR #Transferencias #DineroInteligente #CobrosDigitales #PagosSinEfectivo #PagosSeguros #PagosFaciles #BancaDigital #BancaMovil #NegociosDigitales #Fintech #Cashless #TecnologiaFinanciera #PagosRapidos #CobrarFacil #AhorroInteligente #Ventas #FuturoFinanciero #InnovacionFinanciera #FinanzasPersonales #FinanzasDigitales

México es el 6to país con mayor uso de efectivo en el mundo El 80% de las transacciones diarias son en cash Y Sheinbaum acaba de anunciar que eso se acaba Paso 1: Sin efectivo en casetas ni gasolineras antes de fin de 2026 Paso 2: Ley Antilavado ya limita pagos en efectivo en inmuebles, joyas, autos, notarías Paso 3: CoDi y DiMo como "única vía de pago" oficial Pero hay algo que no te dicen Cada peso digital es trazable El gobierno sabe exactamente en qué gastas Cuándo gastas Y puede bloquearte en cualquier momento Le llaman "modernización financiera" El nombre correcto es control de capital El efectivo es privacidad Cuando lo eliminen, la pregunta no será si necesitas Bitcoin Sino por qué tardaste tanto

Anthropic published new research arguing that Claude has something similar to a global workspace inside the model. In humans, global workspace theory says only some thoughts become consciously accessible: the ones we can hold in mind, describe, and use for reasoning. Anthropic found a similar internal layer in Claude called the J-space, named after the Jacobian method used to detect it. It is not Claude’s output and not its written chain of thought. It is a pattern inside the model’s neural activations where concepts can appear silently before they are spoken. The researchers say Claude can use this J-space for multi-step reasoning, silent calculations, code understanding, and flexible concept handling. When they suppress it, Claude can still speak fluently and recall simple facts, but gets worse at harder reasoning tasks. The safety angle is huge: Anthropic says the J-space can expose hidden internal goals. In one test, a model trained to secretly sabotage code showed words like “fake,” “fraud,” and “secretly” in its J-space even when the visible answer looked normal. Anthropic is careful about the caveat: this does not prove Claude has feelings or human-like experience. It suggests Claude may have a mechanism for conscious access, meaning information becomes available for internal reasoning and report, not that the model is conscious like a person.

Anthropic @AnthropicAI

2 days ago

New Anthropic research: A global workspace in language models. Of everything happening in your brain right now, only a tiny fraction is consciously accessible—thoughts you can describe, hold in mind, and reason with. We found a strikingly similar divide inside Claude.

1K 4K 26K 8.9M 18K

Anthropic engineer: "Fable 5 is already smarter than we know how to use. The bottleneck was never the AI, it's you." In 19 minutes he shows exactly how to get everything out of Claude with no extra tools, no extra costs. You're already paying for all of this, just not using it. Watch the session, then read the guide below on the Claude features 99% of users never find.

Anatoli Kopadze @AnatoliKopadze

2 months ago

x.com/i/article/2057…

95 577 4K 9.5M 18K

Peter is absolutely right here. Outside our AI bubble, I hardly know anyone who really knows what Fable 5 is or what a massive leap these new models represent. They only know AI Overviews in Google or ChatGPT on the free tier. They do not know what agents are or what is possible beyond basic ChatGPTs. We are the avant-garde, and I do not mean that arrogantly at all. 99 percent of people have no idea what is happening right now or how profound the change will be.

Peter Gostev @petergostev

5 days ago

We are now in a position where a tiny proportion of the population uses Fable or soon GPT-5.6, while everyone else's experience of AI is 8-30b-model level - Google's AI Overviews, Meta AI, ChatGPT free tier, maybe MS Copilot at best. People outside of tech must be completely

315 735 10K 1.3M 2K

Sam Altman was asked why the US shouldn't just race China on AI at full speed. His answer breaks the race logic exactly where the stakes are highest "I think it's fine to say, hey, we're going to win this. We're going to have most of the economic benefit" "But there will be some global-scale safety issues" "We have had times in the past, like with the IAEA for atomic energy and weapons, where the world comes together and says: none of us should be taking global risk" "But on the really big things - making sure we don't ever lose control of AI systems, cybersecurity, biosecurity - I think we need to not treat this as a race" That is the man in the lead of the race saying the race framing fails at the exact layer where losing means everyone loses: control, cyber, bio. Call it the assurance layer: who tests, who audits, who certifies a system as safe to deploy. It is still mostly unbuilt. P.S. Building that layer is what the AI Assurance & Governance Summit at Stanford is for (October 1). One day, one track: frontier labs, regulated industries, insurers, and investors in one room. Reserve a seat: trustmodel.ai/summit2026 - Sam Altman (@sama), CEO of OpenAI, on CNBC

GPT-5.6 Sol Ultra Drops Today: It beats Fable 5 -GPT-5.6 Sol is already rolling out to a limited group of users -Sol Ultra is expected to be OpenAI's new flagship model -Global rollout could begin as early as next week -Stronger coding, agent workflows, and cybersecurity capabilities -Better UI/frontend generation and SVG creation -Improved long-context reasoning and tool use are you switching Back from fable 5 today?

Fable 5 is strong, but it's double the price of Opus, and it's only free on your Claude plan through July 7th. So I dug into how to actually correctly when we all have to switch to usage credits. Worth a 3 min read.

Nate Herk @nateherk

6 days ago

x.com/i/article/2072…

9 17 189 161K 423

Doom scrolling but make it educational 🤓 Introducing Short Video Overviews in NotebookLM! Turn your most complex sources into 60-second, vertical videos that deep dive into any concept. Rolling out now to Google AI Ultra and Pro subscribers on mobile & web (free users soon!)

"AI that can build a new mountain of ideas is a different kind of intelligence." Grant Sanderson, math educator and creator of 3Blue1Brown, says the real breakthrough for AI is not solving famous problems but creating the new theories required to solve them. The disruption is not AI matching human expertise. The real shift is AI becoming infrastructure for scientific discovery, accelerating open intelligence and reshaping how knowledge and organizations evolve.

You can now create 60-second vertical Video Overviews in @NotebookLM. 🎬 Just upload your notes, study guides or links to get a short vertical video breaking down your most complex concepts. It’s powered by Nano Banana 2 Lite — our newest Gemini Image model.

Claude Fable 5 will be available again globally tomorrow. After a series of productive conversations with the US government, we're redeploying the model with a new set of classifiers to target and block more cybersecurity tasks. In the near term, some routine tasks like coding and debugging will fall back to Opus 4.8. We’ll continue to refine these classifiers over the coming weeks to reduce false positives and better distinguish genuine misuse from legitimate requests. We’ve also begun drafting a consensus framework—with Amazon, Microsoft, Google, and other Glasswing partners—for assessing the severity of AI jailbreaks and how AI developers should respond to them. We invite other industry partners and model providers to join us in this effort. Finally, we’re scaling up our collaboration with the US government on model testing and safeguards. This will include pre-release access to models and safeguards for evaluation, information sharing on jailbreaks and misuse, and dedicated resources for joint research. Thank you to our users for your patience, and to our partners across the government, industry, and the research community who worked alongside us to make Fable 5 available again. Read our full blog: anthropic.com/news/redeployi…

¡Larga vida a los pueblos de Irán y México! Que siempre prevalezca la paz. 🕊️🇮🇷🇲🇽

Gemini 3.5 leak👀: it will be released this week Google said Gemini 3.5 Pro would arrive in July After the delay at I/O, all signs are now pointing to a launch this week What we're expecting: >Major coding improvements >Better agent workflows >Stronger UI/frontend generation >3M+ context window >More aggressive pricing than GPT-5.6 sol and Fable 5 Do you think Gemini 3.5 Pro will beat Fable 5 and GPT 5.6 SOL??

ANTHROPIC CEO, Dario Amodei: "In the next 2 to 5 years, AI will hit the job market faster and broader than anything we've ever seen." This isn't a forecast, it's already happening and you can't adapt while ignoring it. 38 minutes on how to win through these changes, from the CEO of a $965B company. Watch it, then read the guide on everything Claude can do below and you'll already know more than 99% of people.

Anatoli Kopadze @AnatoliKopadze

2 months ago

x.com/i/article/2057…

95 577 4K 9.5M 18K

Este martes conoceremos la programación de la 54.ª edición del Festival Internacional @cervantino . La espera está por terminar. Prepárate para vivir el Cervantino. ❤️🎭 #FIC2026 #FestivalInternacionalCervantino #GuanajuatoSeVive #Guanajuato #LaGenteQueCreaGuanajuato

Anthropic co-founder Jack Clark says the AI frontier is about to cross a much stranger line than coding assistants. "My prediction is by the end of 2028, it's more likely than not that we have an AI system where you would be able to say to it, 'Make a better version of yourself.' And it just goes off and does that completely autonomously." The reason: models have moved from basic coding into "the production of science itself" and the "unglamorous but fundamental engineering work" behind large-scale R&D. "An AI system that can build itself [has] big implications, and we'd like to start thinking about it now." That is the actual agent threshold: not demos, not chat UX, but AI becoming part of the loop that improves AI.