Tag: AI

AI VFX has come a long way in a very short amount of time and there’s about to breakthrough to being something truly remarkable.

In this video, I outline my workflow and different best practices.

Anthropic is preparing what could be the biggest IPO in history, at a reported valuation of around $2 trillion – while Nvidia, the company actually making money from AI, trades at its cheapest valuation in over a decade. In this video, finance professor Patrick Boyle looks at whether the AI boom is a bubble and what you would have to believe to pay $2 trillion for an AI lab. We cover the Anthropic and OpenAI IPO plans, why rising interest rates and a 5% ten-year Treasury yield are bad news for AI stocks, how “total addressable market” (TAM) estimates have been used to justify tech valuations from Amazon to Uber and WeWork, SoftBank’s junk bonds and its $50 billion data-centre company SB Energy, the circular financing between Nvidia, OpenAI and SoftBank, and the AI price war with Chinese open-weight models like DeepSeek. Along the way, we revisit Scott McNealy’s famous warning about Sun Microsystems and the dotcom bubble, and ask whether today’s AI giants could turn out to be the Yahoo and Lycos of this era.

Kling 4 is here — kinda. But, I ran it through the usual tests: text-to-video, image-to-video, Omni with Flamethrower Girl, a noir scene in Malloy’s office, and my keyframes-plus-Omni trick. Then Google wakes up with Gemini 4 Argon (announced, not released — return of the King? We’ll see), HeyGen drops a video model that starts at a penny a second (asterisk included), Ideogram 4.5 gets precise editing, and ElevenLabs ships Eleven v4. AI fall is getting toasty. WHAT’S IN THIS ONE

• Kling 4.0 early access: what you actually get right now (Flash, 720p, Ultra annual only) and when the full model lands.

• HeyGen Video: HeyGen’s first general-purpose video model, built on MiniMax H3 and post-trained by HeyGen. One cent a second is the 50%-off launch price through the end of October, for 480p text- or image-to-video (two cents normally); 768p is a cent and a half on sale, three cents regular; reference-to-video costs double. API, OpenRouter, Runware and ComfyUI now, web app coming. Haven’t tested it yet

• Gemini 4 Argon: announced, not released (trusted cyber-defense partners first, then paid API and AI Ultra, no date). Google’s own numbers put it ahead of GPT-6 Astra and Opus 5.5 on 12 of the 18 benchmarks it reports, plus a 91.7 on LVBench for long-video understanding. The interesting part is what comes next — Nano Banana 2 Pro, a Genie update, Google back in the video race? And how Google answers Muse and OpenAI’s dots.
• Pour one out for the OG Nano Banana: Gemini 2.5 Flash Image shuts down in the Gemini API on October 2, 2026

• Ideogram 4.5: “the most precise edit model” — edit the same image over and over without the drift. Tested on a ten-month-old image full of garbled text (and a Flamethrower Girl profile-pic swap). Open weights promised. Don’t sleep on Ideogram.

• Eleven v4 and v4 Turbo: natural-language direction right in the script — laughter, sound effects, delivery — tested on the high-fantasy watchtower opening. Available now, including on the free tier.

AI is developing faster than we could have imagined.

What does this mean for future development and how worried should we be?

How far are we from AI developing recursive self-improvement, where AI systems write their own code?

Robinhood Chairman and CEO Vlad Tenev discusses the company’s new AI trading agents, the expansion of 24/7 trading for native equities, and the future of retail investing, on ‘Making Money.’

OpenAI CEO Sam Altman spoke at the UN Security Council meeting about AI and international security on Wednesday.

Paul J Lipsky explores the launch of Anthropic’s latest model, Claude Opus 5.5. The analysis covers updated performance benchmarks, cost efficiency, and improvements in conversational tone compared to previous versions. New changes to usage limits for Pro Max and Team plans are also examined.

Presented as a Star Trek–themed space mission aboard the “Starship Robinhood,” CEO Vlad Tenev and the leadership team announced major product upgrades aimed at expanding market access, trading functionality, and AI automation [15:00], [16:27].

Key Announcements & Upgrades

• 24/7 Stock Trading: Robinhood is upgrading its 24/5 trading market to a 24/7 equity market, bringing native stock trading to weekends [18:18].
• Extended Options Hours & OCO Orders: Options trading hours for popular symbols (NVDA, TSLA, SPY, QQQ) are expanding by 35% (7:30 a.m.–4:15 p.m.) [19:45]. Robinhood is also introducing One-Cancels-the-Other (OCO) orders and 4x intraday margin [20:34], [22:02].
• Robinhood Agents & AI Integration: The headline release is Robinhood Agents, placing AI-driven, hedge-fund-style automation directly inside the app [58:11]. It allows users to build, backtest, and run continuous trading strategies using frontier models [58:54]. Through the Apps Nexus, agents integrate third-party institutional data from providers like Unusual Whales, Quiver Quantitative, and Nasdaq [01:03:17].
• Crypto Perpetuals: Robinhood becomes the first major U.S. brokerage to launch perpetual futures for crypto assets with customizable leverage up to 10x [49:52], [51:16].
• Earnings Predictions & Call Player: Users can now trade company earnings directly via binary prediction contracts [35:42]. An interactive Earnings Call Player was introduced, featuring live synced transcripts, community feeds, and in-app order execution [38:10].
• Robinhood Social & Multi-Display Support: Robinhood Social moves out of beta, featuring verified trading posts and P&L cards [30:03], [31:44]. Additionally, optimized multi-display interfaces are launching for foldables (iPhone Duo, Galaxy Z) and tablets [25:44], [27:27].

I’m excited and worried about the era of AI. We need a plan to maximize the good and minimize the bad.

Making an AI better has meant making it bigger for about six years now, and it has worked well enough that most people treat it as a rule, but a recent paper from MIT found the point where it stops. Every LLM has to hold more concepts than it has room for, so it stacks them into the same directions and lets them overlap. That overlap is called superposition, and it is the reason these models work as well as they do, though it is also the reason they get things wrong. The paper worked out the exchange rate. Double the width of a model and you halve the errors, which turns out to be most of why bigger models have been better all along.

In this video, I break down what a model’s internal space actually looks like, why holding more concepts than it has dimensions is what makes models wrong in the first place, and the three separate things that stop this strategy from running forever, only one of which is money.