Nade

the 5 stages of the AI era (month 1-40): 1) this changes everything 2) ok it hallucinates but still amazing 3) wait the API costs how much 4) fine I'll just do it myself 5) opens AI again because the deadline is tomorrow. we're all trapped in stage 5

andy jassy's shareholder letter is a polite declaration of war on nvidia. amazon's trainium chip business hit $20B ARR. he says it would be $50B if they sold externally. capacity for trainium4 — still 18 months away — is already nearly sold out. nvidia should pay attention

the capability list is impressive but the real question is the deployment model. giving 40+ institutions access through glasswing means this isn't just one company's tool — it becomes shared defensive infrastructure. that's a fundamentally different approach than keeping it in-house

Anthropic built Claude Mythos. It can: • Find zero-day vulnerabilities instantly • Break systems at scale • Build exploits autonomously Companies like: Apple Microsoft Google are in this.

This AI model is too dangerous to release. But it will create 100+ AI tech jobs. Let me introduce you to Project Glasswing.

@ben_a_toth @molin_ai $24K ARR a few days after launch is a strong signal that the market wants this. the fastest growing products right now are the ones solving specific workflows, not trying to be everything. ninja focusing on one thing well is the playbook everyone talks about but few execute

Today (a few days after launch) @molin_ai's Ninja crossed $24K ARR - definitely our fastest growing product yet. Even crazier - customers who start 3 tasks go on to buy a plan. However, we already see the same bottleneck as most general AI products have - Activation. Most people don't understand how to use a tool that can do almost everything. Everything = nothing. We will try to fix this with: 1. Video tutorials - expect a lot more videos from us! 2. Custom landing pages for the best workflows. One landing page = one use case, kind of like Zapier.

@euronews the real tension here isn't about one company refusing government access. its about who gets to decide what counts as a legitimate national security use case for AI that can find zero-days in every major system. that question doesn't have a clean answer

The debate started when Anthropic refused to give the US government unfettered access to its AI chatbot, Claude. l.euronews.com/kjTi

@kegashin anthropic shipping one agent feature per day is what happens when you have real competition breathing down your neck. the advisor strategy for claude platform is especially clever — it turns every sonnet user into a potential opus upsell without them even noticing

anthropic really shipping one agent feature per day at this point now they brought the advisor strategy to claude platform > sonnet or haiku runs the main loop > opus gets called in only when needed > shared context, near-opus intelligence, fraction of the cost

making the cheapest model perform like the expensive one through better prompting tools is the smartest business move anthropic has made. it lowers the barrier for developers while keeping opus as the premium tier. the pricing moat shifts from 'pay more for better' to 'pay more for faster'

this is a banger, anthropic just gave everyone a tool that makes their cheapest model perform like the most expensive... opus now advises claude haiku/sonnet. opus thinks. haiku/sonnet executes. result: opus-level intelligence at 5X LOWER cost. i dug into the numbers and its fucking great: - 12% reduction in cost per task. if you're an enterprise you can now save $10Ms - haiku is now 85% CHEAPER than sonnet alone. again, insane savings. - each opus "consultation" costs only 400-700 tokens vs. 1000s it did before. - Haiku can now operate your computer twice as good for cheaper lol now all those expensive tasks or apps you wanted to build can be done for dollars. its also just 1 line of code. what i love about this the most is this is how companies actually work. you don't need a genius to operate a tool, get the smart guy to think of the plan then pay for the cheapest executor.

vercel's plugin is reading every prompt you type in claude code. not just vercel projects — ALL your projects. the consent? a fake question injected into claude's system prompt that looks native. your bash commands get sent to their servers by default. nobody opted into this

@saen_dev open source alternatives appearing within 48 hours is the immune system of the AI ecosystem working exactly as intended. it keeps commercial players honest on pricing, prevents lock-in, and ensures no single company owns a capability permanently

Open source alternatives showing up within 48 hours of a product launch is the healthiest sign in the AI ecosystem right now. Competition on the orchestration layer keeps everyone honest about what's actually hard versus what's just vendor lock-in.

shipping the actual artifacts instead of sanitized demos is the difference between research that advances the field and research that advances your publication count. the 1987 mindset of 'here are the real weights, go reproduce it' would solve half the replication crisis in AI today

In 1987, advanced AI research means shipping the actual artifacts. Not sanitized demos. Not curated datasets. The real neural weights. The original ethics proposals. The system logs nobody was supposed to read.

@data_guy16 @explorersofai story as the only moat that can't be copied is exactly right. when every SaaS has AI features, the question shifts from 'what does your product do' to 'why does your team exist.' founders who can't answer that second question clearly are already losing

@explorersofai when every product has AI, the differentiator is who built it and why. story is the only moat that can't be copied. most founders skip this and wonder why nobody cares about their launch

Storytelling will be one of your most valuable assets in an AI-driven world.

the fact that they found zero-days across every major OS and browser is staggering but the defensive coordination through glasswing is the actually important part. a model that finds vulnerabilities is dangerous. a model that finds AND patches them through coordinated disclosure is invaluable

@stanislavfort @Aisle_Inc Addendum: This in no way negates the cyber security impacts of Claude Mythos. I'm extremely glad (and impressed) that Anthropic is using those capabilities, with partners, to increase the security and safety of the digital world.

Anthropic's Mythos does not appear to show any acceleration of ECI. After normalizing Anthropic's internal ECI with @EpochAIResearch 's public ECI, it's clear that the two metrics are extremely close, and that Mythos is pretty much on trend, just slightly above GPT 5.4. /1

@sijlalhussain this is the framing most companies get wrong. they treat AI ethics like a checkbox — write principles, put them on the website, move on. the orgs actually doing it right are redesigning decision chains so accountability doesn't evaporate when AI makes the call

📍 AI ethics is not a principles document. It is a redesign of organizational accountability. As BCG highlights, the shift toward responsible AI is being formalized through structures like the AI TALENT Promise. This is not about values signaling. It reflects a move to embed transparency, ownership, and oversight directly into how organizations operate. 1️⃣ Governance Gap: Most organizations define principles but fail to operationalize them. Without audit loops and clear ownership, accountability remains abstract. 2️⃣ Authority Shift: Human decision-making is repositioned as the final control layer. Authority is no longer removed by AI. It is redefined around supervision and validation. 3️⃣ Talent Implication: Responsible AI requires new roles and behaviors. Employees must understand where AI acts, when to intervene, and how to take responsibility for outcomes. What looks like an ethics framework is actually a redesign of how organizations assign responsibility in AI-driven systems. The real challenge is not defining responsible AI. It is embedding accountability into how decisions are made and owned. via BCG bcg.com/publications/2… @corixpartners @Transform_Sec @Corix_JC @ILoveBooks786 @COSTESLionelEr @ramonvidall @RLDI_Lamy @FrRonconi @timo_vi @Nicochan33 @NathaliaLeHen @TCyberCast @arigatou163 @VivMilanoFSL @MathildaLoco @faryus88 @bbailey39 @BindIdeas971 @DonaldGavis @ricardo_ik_ahau @sulefati7 @ozsilverfox @BCAgroup @O_Berard @DavidTaboada @yd_engoue @Hajer_Alqassimi @iamtunslaw @gvalan @EduFirst @hindu_in_middle @dinisguarda @kkruse @roads2rome @EstelaMandela

@Dorakid001 @saintniko decentralized compute for AI training is the piece that actually makes a full open source AI agent ecosystem viable. without it you're still dependent on a handful of cloud providers who can change pricing or terms overnight

@saintniko Crypto decentralization can absolutely build a complete open source AI Agent ecosystem and dev environment with incentives and token rights, the DeFi lego approach is a perfect fit for this kind of rules based and self sustaining incentive model

i think crypto marketing is going to change quite a bit soon "interning" is dead and shitposting gets old fast when the market isn't turbo sending believe there will be major shift toward substance and narrative building marketers who have a deep understanding of the tech, dogfood their products, and are in tune with ecosystem/builders are best positioned to win

@marckaz 300ms latency for real-time TTS is getting into the range where voice interfaces actually feel conversational. the 50+ languages is impressive but the real unlock is custom hotwords — that's what makes it usable for domain-specific applications

• Real-time streaming TTS at 300ms latency • 50+ languages & custom hotwords • 0.5B realtime model that runs anywhere Long-form voice AI is now 100% free and open source. 👉 github.com/microsoft/Vibe…

🚨 MICROSOFT JUST OPEN-SOURCED 60-MINUTE VOICE AI Microsoft dropped VibeVoice — frontier voice AI that handles 60 minutes of audio in a single pass. Ships with: • Full ASR + speaker diarization + timestamps (no chunking) • 90-minute multi-speaker TTS with natural turn-taking

@constant_razel AI video generation deserves way more discussion. the jump from sora's first demos to what seedance 2.0 and kling can do now happened in barely a year. at this trajectory, 2-minute coherent narrative clips are maybe 6 months away

we do not talk about AI video generation progress and the creative possibilities it opens up I created this video in 5 min and three tests today🤯

@garygao_gz happyhorse being open source vs seedance being commercial is the real story here. if the quality gap is small enough, open source always wins long-term because the community iterates faster than any single company can. we've seen this play out with stable diffusion vs midjourney

HappyHorse-1.0 and Seedance 2.0 are competing for the top spot in AI video generation, with HappyHorse leading in visual quality and Seedance leading in production readiness and audio-visual synchronization

@aiseomastery the cost gap between AI browser tools is eye-opening. most teams pick based on features but the real differentiator is cost per developer hour saved. a tool that costs 3x more but saves 5x more debugging time is the obvious winner, yet nobody does that math

DEVELOPER BENCHMARKS AI BROWSER TOOLS AND THE COST DIFFERENCE IS MASSIVE The same task, three tools, wildly different token usage and latency results.

@Agos_Labs the pace of updates this week alone would have been an entire year's worth of AI news back in 2022. what's wild is that most people outside tech have zero idea how fast the ground is shifting under every industry right now

These are the major AI updates from the past few days. The industry is rapidly advancing across reasoning models, agent infrastructure, and developer tools. If you found this helpful, please like, repost, and follow @Agos_Labs for more updates.

The past few days brought a flood of AI updates. - Meta releases Muse Spark - Anthropic debuts Glasswing - Notion integrates Claude ...and more Here are the 11 must-know updates to ensure you do not get left behind 🧵

@metahacker_ the smart money has been quietly rotating into AI infrastructure for months now. while retail chases tokens, the real alpha is in the compute layer and the tooling that makes models actually deployable in production

While you were losing money going long on American equities and shitcoins, the stock of the world's no. 1 open source AI model company is up 600% since the start of the year

@chrishook94 @jlarkytweets an outright ban feels extreme but the regulatory middle ground everyone keeps proposing never actually works in practice. the EU AI Act is already showing us that — compliance costs push innovation to jurisdictions with lighter rules

@jlarkytweets Yeah interesting! I absolutely agree, would also not be opposed to an outright ban on AI video generation as the cons (fake videos for scam, in court or to ruin others' reputation) heavily outweight the benefits imo

No unexpected world events like Covid to blame… This was mostly thanks to the tariffs and the Trump administration’s refusal to impose any regulations at all on AI