@jun_song “Still using Gemini?” Apparently ~900M monthly users, 2.5B using AI Overviews, 16B API tokens/min, and enterprise demand so high Google is rationing capacity. But sure, your timeline is a flawless market study.
@jun_song “Still using Gemini?” Apparently ~900M monthly users, 2.5B using AI Overviews, 16B API tokens/min, and enterprise demand so high Google is rationing capacity. But sure, your timeline is a flawless market study.
GOOGLE FACES AI CAPACITY CRUNCH AS DEMAND OUTSTRIPS SUPPLY: REPORT
Google is reportedly struggling to keep up with surging demand for its AI services due to limited computing capacity. The supply constraints highlight the intense competition for AI infrastructure, with tech companies racing to expand data centers and secure high-performance chips to support growing enterprise and consumer adoption.
Not as relevant now :-(: I had an opportunity to deeply test both Fable 5 and GPT-5.6 Max. 5.6 is clearly better than Opus 4.8 at everything (slightly faster, too, though that depends on the load). Vis-a-vie Fable, it is clearly worse on coding, but better on agentic workloads. I had Fable write code, 5.6 run experiments - dreamy…
@lily_gpupoor@devindesktop@dabit3@swyx@ZixuanLi_ GLM Coding Max x20 is trash: the 5-hour, daily, and weekly token limits run out way faster than with Codex/Claude Code. I don’t recommend it at all until they release something like Fable 5 or better.
@lily_gpupoor@devindesktop@dabit3@swyx@ZixuanLi_ GLM Coding Max x20 is trash: the 5-hour, daily, and weekly token limits run out way faster than with Codex/Claude Code. I don’t recommend it at all until they release something like Fable 5 or better.
Ex-Google engineer explained AI agent loops, harness, evals in 20 minutes - better than 500$ courses.
trace every run → judge it with an LLM → diagnose → fix → ship.
That loop is how agents self-improve over time.
Agent loops + memory + harness + evals - thats the stack.
Watch it, then save the framework below.
Gemini 3.5 Pro v3 Three.js clock output (not final). v3 is down right now for updates, but look at the depth. Simple look, but the inner mechanics are incredibly detailed. Way better than v1 & v2. Waiting for the next checkpoints!
American AI labs need to be protected from the Epstein administration.
The real question is whether they can find a way to serve global demand through trusted international hosting/partnerships, while keeping frontier research, model weights, and core compute in the US.
As I’ve been saying: Google likely has AI at or above OpenAI/Anthropic in science & math, though not coding. But that isn’t its business model. With models like Flash 3.5 it can barely keep up, so it won’t offer anything better unless it can afford it at billions-user scale.
🚨BREAKINGG : Jun Song, Ambassador @ Alibaba_Qwen, just verified that "A new Chinese AI model from Zhipu AI reportedly matches Claude Mythos’ performance at finding security bugs." is coming soon.
Some of the most important ideas in physics begin with mathematics, and null space is one of them.
In physics, equations are often written as systems of linear equations. The null space contains all solutions that produce zero output under a given linear transformation. It plays a central role in quantum mechanics, electromagnetism, relativity, and field theory, where it helps identify conserved quantities, symmetries, gauge freedoms, and physically equivalent states.
The mathematical concept of null space is well established. However, the image also includes claims linking it to superluminal regions, special vacuum properties, and other speculative ideas that are not supported by mainstream physics. The genuine role of null space lies in the mathematics that underpins modern physics, not in these unverified interpretations.