Overview
Speed and pragmatism ran through today. China is pushing EV charging to near absurd levels, Europe’s small launchers are warming up for orbit, and AI feels much more hands-on, from background dev servers to agents that handle email and tasks. Creative tools got simpler, robots did the chores, and the AGI debate stayed lively. Culture-wise, “chopped” took the crown, while a 1980s trunk of cash reminded us that truth is stranger than fiction.
The big picture
BYD’s 5-minute, 400 km charge raises hopes and hard questions
Dominic Lee shared a dashboard-side video of BYD’s 1000 kW system, with power peaking at 746 kW and range jumping from 135 km to 739 km in minutes, easily outpacing Tesla’s V4 rate. Replies point to grid strain on par with hundreds of homes per session and to battery wear, though independent tests credit BYD’s Blade packs with better heat control.
QVAC’s Genesis V2 backs “option-level reasoning” with data
The team opened up a 148B-token education dataset across 19 domains and introduced a method that explains why distractors are wrong, not just which choice is right. Benchmarks show ~30% average accuracy vs ~12% peers, plus 99.4% valid, clear answers, and a short demo to boot.
YC: AI moved from chaos to “buildable”, with apps in the spotlight
Founders in the latest YC batch favour Anthropic for coding and agent tasks, with Gemini used by nearly a quarter for grounded information. The centre of gravity is now the application layer, where teams orchestrate multiple LLMs, treat intelligence as a commodity input, and iterate faster. Hiring remains a choke point, and domain-tuned models look ascendant.
A 1984 airport trunk held cash, gold and platinum
Dr. Lemma resurfaced the SFO find: $5.64m in cash, 450 maple leaf coins, and 500 platinum bars. A judge sent the full $6.22m, after metals were auctioned, to the US government under civil forfeiture. Two Budget staffers later received $75k each, roughly $214k in today’s money, though taxes bit hard.
Robots make sandwiches and scrub pans after under 9 hours of fine-tuning
Physical Intelligence’s π0.6 helped robots unlock doors, wash greasy pans and build sandwiches, hitting 52% average success across five “Robot Olympics” tasks. The work leans into Moravec’s paradox, showing why sensorimotor skill remains tough even as reasoning improves, and why physical pre-training matters.
Isar’s Spectrum clears final static fires for a January orbital shot
Both stages passed 30-second integrated tests at Andøya, less than nine months after the first test flight. The Arctic footage, aurora included, tees up an orbital attempt carrying ESA cubesats, roughly 150 kg in total.
Lovable now learns how to use any NPM package
Anton Osika showed an AI-built analytics dashboard with 12 components deployed off a single npm install inside chat. The $6.6b startup continues to let prompts become production-grade apps, drawing cheers from both coders and non-coders.
Hassabis vs LeCun, round next: what counts as “general”
Demis Hassabis pushed back on the idea that general intelligence is an illusion, arguing brains are highly general, near Turing-machine-like. The thread nods to real experiments with JEPA on Jetson boards, showing how theory and practice prod each other forward.
Shapr3D models a spring-loaded belt scraper on iPad
A brisk 48-second time-lapse runs through sketches, revolves and assemblies to design a scraper that clears sticky residue without harming the belt. Engineers praised the speed and touch-first workflow, while quizzing features like auto-threading and desktop parity.
Codex gets background processes for uninterrupted dev work
Greg Brockman previewed terminals that keep long-running tasks, like a Vite server on localhost:1420, going in the background. Developers liked the reduced friction, though some asked for stronger process controls compared with tools like Cursor.
Manus AI claims $100m ARR in 8 months
hidecloud recapped an eye-watering growth run, with two products in a16z’s top 50 list and claims of beating frontier models on key tests. Reports and the company’s blog back the revenue milestone as record-fast.
Claude drafts a mansion in Unreal, no manual coding
Claire Silver demoed UnrealMCP, a free plugin that lets an AI drive Unreal Engine via plain English, building a modern mansion from primitives on command. The mini tutorial sparked fresh excitement about AI-first level design and rapid prototyping.
“Starfleet Academy” as a real plan, not just a meme
Elon Musk’s clip frames interstellar travel and the hunt for ancient alien ruins as the next frontier, noting the ages of the universe, the Earth, and human civilisation. The timing tracks with Starship progress and the ongoing case for a multiplanet species.
Altman’s take: Google’s mindset lost, not its people
Sam Altman argues bolt-on AI cannot beat full reinvention. Debate ensued, with Gemini 2.0 and Google’s market cap put forward by sceptics. OpenAI’s surging API revenue suggests a different path, though research cautions that hybrid tactics often give mixed outcomes.
Grok Imagine templates make instant holiday videos
Pick “Let it Snow”, upload a photo, and get a video without prompts or edits, as shown in a breezy screen recording. The light lift invites playful experiments and shareable clips.
X’s Word of 2025: “chopped”
The site crowned “chopped” after 4.1m uses, well ahead of “aura farming” and “vibe coding”. The revived slang has morphed into a jokey way to call a look or moment ugly, and replies ran with it.
ClickUp’s Super Agents pitch human-like skills at work
Zeb Evans introduced agents that DM, take mentions, send emails and manage tasks with user-level permissions, built in minutes via a prompt-led studio. The bet is that context-aware helpers can cut the enterprise AI failure rate, though the real exam lies outside ClickUp’s own garden.
Why it matters
Infrastructure meets ambition. BYD’s 1 MW charging promises pit stops that feel like petrol, yet they also expose grid limits and battery care trade offs. Countries racing to EV scale will need thicker wires, smarter load balancing and honest warranty terms.
AI is getting hands-on. Background processes in Codex, Lovable’s NPM know-how, and ClickUp’s agents turn AI from chat to helper and colleague. YC’s read is spot on, the action is in orchestration, permissions, and product fit, not only in the next giant model.
Embodiment inches forward. Robots at 52% success are not housekeepers yet, but the fast gains from modest fine-tuning hint at a near future where “try, learn, try again” brings real utility in homes and warehouses.
Creation is being simplified. UnrealMCP and Grok’s templates drop the bar to ship a scene or a seasonal clip. Expect more people to prototype, test, and post, which raises the bar for originality while widening the pool of creators.
Space and story still inspire. Isar’s progress gives Europe another route to orbit, and Musk’s Starfleet talk keeps eyes up. Culture keeps pace too, from “chopped” to the SFO windfall tale, reminding us that tech lives alongside language and law, not apart from them.
