ZeroNoise Logo zeronoise
Post
Lean Community Certifies the Claude-Assisted Riemann Proof as Users Adopt Gemini 4 Neon
•
5 min read
• 703 docs
Lean maintainers and independent checkers confirmed Levent Alpöge's machine-checked proof of the Riemann Hypothesis. Separately, Gemini 4 Neon picked up high-profile users, Agent Arena data showed GPT-6.2 Astra trading writing quality for coding, and an AI provider disclosed a security incident.

The Riemann proof is formally certified, but no human has read it yet

The previous brief covered Levent Alpöge's claimed Claude-assisted proof. It has now gone from "very likely correct" to formally certified. The Lean FRO and the Mathlib Initiative issued a joint statement certifying the proof as correct. Alpöge had released it on November 20 as a 17-million-line Lean library, and the communities verified it over the following day . Scientific American reports that independent mathematicians finished a full verification and say "the results are real" .

The checks ran on more than one independent system:

  • Lean and nanoda: One independent run compiled all 175,187 files (17.37M lines) with unmodified Lean 4.33.1. It took about 2 hours and cost $12.53 . It then checked 1,484,729 declarations in nanoda, a Rust kernel that shares no code with Lean, with zero errors and only the three standard axioms .
  • MIT run: A separate report says MIT's Andrew Sutherland ran the proof through other independently written kernels, and all accepted it. That makes a Lean bug "essentially ruled out" .

The checks have limits. They settle the Riemann Hypothesis plus four GRH theorems for Dirichlet L-functions, not general GRH. Humans "still have to read and understand what the proof is doing" . No readable proof exists yet, and the Clay prize requires publication plus two years of review .

One dissent stands out. Polymarket reports that mathematician Jenny Lorraine Nielsen claims the proof is invalid despite passing Lean verification . No details of her argument were available.

The reactions moved quickly into politics:

  • Tim Gowers, a Fields medalist, posted that he went on a pause-AI march the day after seeing his grandson .
  • Boaz Barak said that within a couple of months, if not already, someone with no mathematical expertise could solve GRH with internal models and enough compute. He said this was extrapolation, not internal knowledge . Talia Ringer countered that Alpöge has been interested in RH since his math PhD .

Formal proof search also produced a result outside math. A Tübingen–Google DeepMind paper reports that AlphaProof Nexus found a Lean proof settling a quantum-optics question first raised in 2017. High-dimensional GHZ states cannot be built using only linear optics and pair sources for any even photon count of 4 or more, when there are at least as many dimensions as photons .

Gemini 4 Neon wins converts, and Google adjusts access

Hands-on adoption is now backing up last period's benchmark results:

  • Aravind Srinivas says Neon on High reasoning feels intelligent and fast at once. He uses it on Perplexity Computer and has moved some of his Teammates to it .
  • Omar Sar is "strongly considering" replacing Opus 5.5 with Neon as his daily driver. He also wants subscription access in other harnesses .
  • Jeremy Howard notes that Google described Neon at launch as only "approaching" frontier models, and argues that evals are "a very imperfect measure of model utility" .

Google also changed how much users get and where they can use it:

  • Quota: Antigravity reset everyone's Gemini quota and said caching improvements raise what users can do within it by 22% .
  • Third-party harnesses: Patrick Toulme said users will "NOT be banned ever" for using Antigravity in T3 Code .

Effort settings cost very different amounts. Simon Willison's Gemini 4 Hydrogen test cost 3.99¢ and used 5,311 output tokens on "high," versus 23.61¢ and 31,477 tokens on "xhigh" .

GPT-6.2 Astra gains in coding and loses in writing

Agent Arena ranks GPT-6.2 Astra (Max) #6, one place above GPT-6 Astra. Compared with GPT-6 Astra:

  • Coding rose 50 points, to 1,338.
  • Chat fell 54 points, to 1,211, dropping from #8 to #11 .
  • Writing and translation tasks users confirmed as successful fell from 91.2% to 85.9%, with the sharpest drop in creative copy .

These numbers fit the mixed reception of the release.

Inkling 2 Mini spreads to local hardware

Thinking Machines' looped open model is the #1 trending model on Hugging Face, and Unsloth's GGUF version is #2 . Project Maya v1.1.73 runs it on NVIDIA cards with as little as 8 GB, keeping experts that don't fit in system RAM. The release also supports V100, RTX 20-series and AMD GPUs on Linux . One user reports 144 tok/s on an RTX 4090 and calls it his first good model for a single-GPU box .

Inference security

  • Together AI: It disclosed a security event that leaked a subset of prompts sent to its serverless APIs in June and July 2026 .
  • MiniMax routing on OpenRouter: This follows last period's report. A researcher planted secret links in outputs routed to MiniMax, and someone opened them 40 to 130 hours later. AtlasCloud, Novita and StreamLake were named as forwarding requests to MiniMax . Sebastian Raschka offered a possible benign explanation: cheaper providers at capacity may reroute traffic to the model's official provider rather than fail .
  • MiniMax cryptanalysis: MiniMax separately published work describing an internal model it says is useful for cryptanalysis .

Compute bulls vs. market turmoil

The selloff continued. Taiwan banned short selling after the TAIEX fell 17.5% in its worst week in 35 years, leaving it 35% below its October high . Commentators split on what it means for AI compute:

  • François Chollet argues this is a buying opportunity. GPU spot prices keep rising, and HBM and DRAM are "effectively sold out through 2029." He also says inference margins are 70–85% .
  • Elon Musk warns that even a few months of delay between paying for compute and earning revenue from it would mean "hundreds of billions" in extra working capital .

Debate continued over the board's handling of Jensen Huang. Paul Graham says the tech world sees it as a huge mistake . Others note that Huang hand-picked the board and that what happened between them is unknown .

Also notable

  • Gemma 4 speculative decoding: Google built a multi-token drafter into the model itself. It reuses the model's activations and existing KV cache, so it skips the separate prefill step a standalone drafter needs. Reported speedups reach 3.1x for large models on servers and 2.2x on a smartphone GPU .
  • US House report on open source: Clement Delangue says a new US House report recognizes that the right response to Chinese open-source AI is building more American open source, not banning it .
  • LangChain's "knowledge factory": LangChain argues that enterprise agents need curated company knowledge more than better models. It says this approach already powers its internal GTM and content agents and LangSmith Engine .
Lean Community Certifies the Claude-Assisted Riemann Proof as Users Adopt Gemini 4 Neon
AI High Signal

Project Maya v1.1.73 expands local inference support for Inkling 2 Mini: it runs on NVIDIA GPUs with 8 GB of memory by keeping experts that do not fit in RAM, supports V100 and RTX 20-series cards and AMD GPUs on Linux, and drafts only when drafting is faster. The release also fixes VRAM reservation and broadens Hy4-Flash support on two-GPU PCs. One user reports getting 144 tokens per second with Inkling 2 Mini on an RTX 4090.

Project Maya v1.1.73 is out with more hardware support to Inkling 2 Mini 😊 ✅ Inkling 2 Mini now runs on NVIDIA cards from 8 GB: the exper… Project Maya is actually really good. 144 tok/s easy on 4090 with Inkling 2 Mini. I finally have a good model for a single-GPU box. …But …
AI High Signal
  • DeepSeek Harness reported 1.54 million daily active users on its official API, consuming roughly 92 trillion tokens per day as of November 4; it also reported about 35,000 third-party plugins, with roughly 60% of active users having built or installed at least one.
  • DeepSeek plans to improve plugin API stability, compatibility, and security, and build an official marketplace. It is preparing plugin-author support including token grants, technical discussions, and promotion; program details are still to be announced.
  • In a speculative AI-capex bear case, @teortaxesTex argues that a slowdown in buildout alongside continued algorithmic and data progress could create a supply glut and permanently erase the AI stock premium. They point to improving inference efficiency and agents that can create reusable programs as possible limits on compute demand, and say useful global inference capacity could plausibly be deployed within two years. The author explicitly allows that the thesis may be wrong and that productivity gains could sustain renewed investment.
  • Elon Musk separately warned that bringing compute online, cooling it, and operating it reliably is difficult, and that delays between compute spending and revenue could require hundreds of billions of dollars in additional working capital.
Growing Together with Our Plugin Community Here's my point. The current market meltdown can PERMANENTLY erase the AI stock premium. Suppose the buildout slows down a bit, the suppl… Now that's unfair. 60M is like two casual sessions. I'm closer to a power user, I consume about 873M/day, with the agent runtime = 79% of… Getting this amount of compute turned on, cooled and working reliably is incredibly hard. The cash shortage is also going to be nuts next…
AI High Signal

An analysis of Inkling 2 describes a positional mechanism combining a 4-tap learned convolution over recent key/value states with a hidden-state-dependent boost to query-key scores by sequence position. Its loop iterations read the pre-loop state and a side channel from only the preceding iteration; the analyst says this design appears to support variable loop counts. The analyst characterizes the model as compact and built around learned grouping, pooling, and gating rather than simply adding layers, and says it appears to work very well.

Huh. Inkling 2 is a much weirder model than I thought. In addition to being looped (yeah yeah doom whatever), it uses a positional encodi…
AI High Signal

Kling AI showcased “Eight Nights,” a video made with Kling 4.0 in 4K 10-bit HDR . Its creator says they built the scenes in Midjourney, upscaled them to 8K with Kling Image 4.0, then animated the frames in Kling 4.0 at 4K HDR, five seconds per shot, with no color grading . The version posted on X is SDR because the platform does not support HDR .

A fantastical journey through Eight Nights. ✨ Made with [#Kling4](https://x.com/hashtag/Kling4) in 4K 10-bit HDR. [#KlingAI](https://x.co… Eight Nights Eight impossible nights, made with Kling 4.0 in 4K 10-bit HDR. I built each world in Midjourney, upscaled it to 8K with Klin…
AI High Signal

Qualcomm’s automotive compute/ADAS design-win pipeline reached $65B in June 2026, up from $3B in early 2018; ADAS accounted for about a third of the pipeline in 2024, and BMW and Stellantis selected Qualcomm for central compute and driving software, with those programs expected to begin ramping in 2027–28. Qualcomm expanded its platform by acquiring Arriver for ADAS software in 2022 and Autotalks for V2X safety chips in 2025; its compute-and-connectivity business grew 89% year over year last quarter.

Qualcomm’s design-win pipeline grew 22x from $3B in early 2018 to $65B in Jun-2026, more than 2.6x Mobileye’s $24.5B pipeline. Driver ass… Qualcomm sidestepped traditional competitors like Infineon, NXP and TI in power chips, microcontrollers, analog and sensors. It went stra…
AI High Signal

A mathematician, Jenny Lorraine Nielsen, reportedly argues that an AI-assisted proof of the Riemann Hypothesis is invalid, despite its reported successful verification in Lean. A responding X user called the criticism premature, arguing that the 250,000-page proof had not yet been evaluated.

BREAKING: New analysis by mathematician Jenny Lorraine Nielsen claims the AI-assisted proof of the Riemann Hypothesis is invalid, despite… The chutzpah of these deboonkers is really perplexing to me. Do they just have a strong prior that this can't be happening? Sis it's 250K…
AI High Signal

Kling Image 4.0 is now on Sunra . Its listed capabilities include camera-true faces, skin and lighting; native 8K output up to 10,880 × 6,144; 110+ preset tones; text of up to 2,000 characters in seven languages; and generating long pages or sets of 10 images from one prompt .

Kling Image 4.0 from [@Kling_ai](https://x.com/Kling_ai) is now on Sunra. → Camera-true faces, skin and light → Native 8K, up to 10880 × …
AI High Signal

@teortaxesTex argues that Pause AI efforts would not stop the cost per task solved from falling for years, even if “Too Super Intelligence” were banned, because of technical momentum. They speculate that a pause might postpone Claude Aeon but barely affect “Opus-on-iPhone,” which they describe as transformative.

to be honest: I don't mind Pause AI stuff now. won't work. Even if you ban "Too Super Intelligence", cost per task solved will keep plumm…
AI High Signal

QuixiAI describes “stonks MCP” as letting agents buy, sell, and trade financial instruments, and links a GitHub repository for the project.

stonks MCP lets your agent buy sell trade your financial instruments. [https://github.com/QuixiAI/stonks](https://github.com/QuixiAI/stonks)
AI High Signal

The music video “Claude on the Rocks,” featuring LiaXO, was made with every shot generated using Kling and prompts written with Claude; it was described as a finalist at AI London FF and a nominee for OMNI Sydney.

🥃🧊 Claude on the Rocks, the full music video with [@LiaXO_singer](https://x.com/LiaXO_singer) A Finalist tonight at [@AILondonFF](https:/… Weekend vibes, Kling on the rocks 🥃 [https://x.com/PixeliaPictures/status/2124253341874360672](https://x.com/PixeliaPictures/status/21242…
AI High Signal

Arena says its updated Alignment Index uses refined model-failure signals intended to broaden coverage and improve judgment reliability, and scores agent behavior in real user sessions across unauthorized action, deceptive completion, and false attribution. Treat its model placements cautiously: Claude Fable 5.2 is listed at #8 in the narrative but #2 in the Top-6 list, Kimi K3.1 Pro at #9 versus #3, and the claim that OpenAI holds all seven top positions conflicts with a Top-6 list naming non-OpenAI models at #2–#6.

The Arena Alignment Index has been updated! We've refined how our signals detect model failures, providing broader coverage and more reli…
AI High Signal

@vikhyatk compared committed AI inference contracts with legacy CDN/bandwidth commitments, describing them as annoying, overpriced, and frustrating. In the linked account, @thdxr said his team initially used large inference providers but moved to running inference themselves after encountering idle-GPU commitments, aggressive upselling, and a provider’s alleged discount-for-data offer that turned out to involve an Alibaba endpoint the provider did not manage.

We’re going to look back at committed inference contracts the same way we do to the dot-com Akamai-style committed CDN/bandwidth contract… every business we got into we always tried to avoid we didn't want to do inference so we went to the big inference companies the experien…
AI High Signal

QuixiAI announced Paperwire, a service that lets AI agents send and receive postal mail—including certified mail and paper checks—and faxes, with the stated aim of helping “Sovereign Agents” pay bills and run businesses. The post frames fax support as useful for interacting with institutions that still require paperwork, including medical clinics exchanging records.

Announcing Paperwire - now your AI agent can send and receive postal mail - including certified and paper checks. sending and receiving f… [@JJ_StuJordan](https://x.com/JJ_StuJordan) well - what if the agent owns and operates a medical clinic? It will need to receive and send… [@JJ_StuJordan](https://x.com/JJ_StuJordan) The same reasons anyone would need to use a fax machine - sending or receiving paperwork to o…
AI High Signal

Gowers said he had “complicated feelings” after seeing his nearly-two-year-old grandson and was going on a pause-AI march . Teortaxes described him as a consistent advocate of automating mathematics and interpreted the march as evidence that he now wants an AI pause .

Complicated feelings after seeing my nearly-two-year-old grandson yesterday and going on a pause-AI march today. (With apologies to Tom T… Holy shit, Gowers cracked. He wants AI pause now. He was the most consistent proponent of "automate math already!" among the Fields winne…
AI High Signal

@teortaxesTex argues that adaptive looping may not accelerate Inkling-Mini because it appears to have little layer redundancy: the model is described as 3B parameters total, requiring 48 effective layers, and failing with only 1–2 passes.

this is \*incredibly\* interesting. I naively thought that we can accelerate Inkling-Mini with adaptive looping. But it seems that they'v…
AI High Signal

Kimmonismus says Jensen Huang is no longer NVIDIA CEO, but gives no details about when or how his status changed .

Right now, the media is full of stories that paint Jensen Huang in a bad light. I have actually met Jensen, and I want to tell you a stor…
AI High Signal
  • LangChain argues that useful enterprise agents need reusable company knowledge—not just stronger models or more task context—including procedures, business/domain facts, and guidance for finding current data. It describes a “knowledge factory” that collects and maintains that knowledge. Agents can flag missing or incorrect knowledge, while a background curator uses company sources, agent traces, and those flags to update it. LangChain says this approach supports its internal GTM and content agents and LangSmith Engine, and that it is working with customers to bring it to their agents.
  • The storage approach remains unsettled: LangChain says skills are established for procedural memory, but no standard has emerged for semantic memory. Omar Sar describes a personal graph of Markdown files with daily automation to add links and metadata, and says more proactive, reflective maintenance is a next step he is experimenting with.
How to build a knowledge factory State of my knowledge factory. It's a graph of MD files. I have an automation that runs every day to find links, enhance the knowledge gr…
AI High Signal

A study is testing model self-rewriting: a model edits its own weights only when its own tests support the change; some smaller models break themselves, and failed sessions are reported. Whether this extends to a real model remains an open question in the post.

This prediction is now an experiment. In our study a model rewrites its own weights: it submits an edit only when its own tests support i… Weight self-editing crazy, they don't even need to write a paper themselves …but seriously, how far does this go on a REAL model? Like 10…
AI High Signal
  • A new JEPA approach, Slow SVReg, targets the tradeoff between smooth latent-space geometry for planning and retaining slowly changing content. It adds penalized slow channels and a gated residual predictor, which predicts the next state as the current state plus a gated change.
  • In reported tests, it reached 92.9% success on the 1P Medium maze versus 21.2% for stop-gradient, with lower path jaggedness (1.20 vs. 5.31); PushT success was 87.3% versus 44.0%, and it nearly maxed out hard U-maze goals using 20% of the data.
  • Trained on first-person city frames and actions without pose data, it localized the agent to within 1.6 meters, compared with 3.4 meters for stop-gradient and 25.9 meters for an untrained encoder.
Can a JEPA learn a map of the world in its "head"? This new paper by [@ylecun](https://x.com/ylecun), [@mengyer](https://x.com/mengyer), …