# Alex Korbonits Personal blog and portfolio. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## OpenAI's Claimed Navier-Stokes Result Covered Alternatives Fefferman Included in the Official Problem DevFeed: [OpenAI's Claimed Navier-Stokes Result Covered Alternatives Fefferman Included in the Official Problem]() Original publisher: [Read original article]() Published: 2026-09-08T00:00:00Z Content type: opinion Language: en Sources: [Alex Korbonits]() Topics: [OpenAI](), [Mathematics](), [Math and Logic]() Tags: [mathematics](), [openai](), [statement]() ### AI overview The article argues that OpenAI did not prove the standard Navier-Stokes existence and smoothness problem, but did prove alternatives (C) and (D) included in Charles Fefferman's official 2000 problem statement. It distinguishes that result from the unresolved alternatives (A) and (B), which concern existence and smoothness on ℝ³ and the torus with zero force. ### Source excerpt OpenAI did not prove Navier-Stokes, and it also did not solve the wrong problem. It proved alternatives (C) and (D), which Fefferman put in the official statement on purpose in 2000, using theorem statements DeepMind had already written. ## Anthropic Formalized Fermat's Last Theorem in Lean, Shifting the Verification Challenge DevFeed: [Anthropic Formalized Fermat's Last Theorem in Lean, Shifting the Verification Challenge]() Original publisher: [Read original article]() Published: 2026-09-06T00:00:00Z Content type: opinion Language: en Sources: [Alex Korbonits]() Topics: [Lean](), [anthropic](), [Math and Logic]() Tags: [anthropic](), [compiler](), [formalized](), [research](), [statement](), [trust]() ### AI overview The article argues that Anthropic's machine-generated Lean proof of Fermat's Last Theorem changes the challenge from formalizing the theorem's statement to checking a very large proof artifact. It describes public build checks and an independent Rust-based Lean kernel re-check, while noting that the author has not read the patches used to complete that re-check. ### Source excerpt Anthropic formalized Fermat's Last Theorem in Lean in eleven days. It is the frontier I said in May was untouched, and the first result in this series where nobody had to trust the statement. The problem that replaced it is that 13 million lines is more than anyone can read. ## OpenAI's Ten Mathematical Results Tested Through Lean Certificates DevFeed: [OpenAI's Ten Mathematical Results Tested Through Lean Certificates]() Original publisher: [Read original article]() Published: 2026-08-01T00:00:00Z Content type: article Language: en Sources: [Alex Korbonits]() Topics: [Lean](), [OpenAI](), [certificates](), [trust](), [Artificial Intelligence]() Tags: [ai](), [certificates](), [openai](), [paper](), [trust]() ### AI overview The article examines OpenAI's release of ten results on long-standing mathematical problems and reports independently building and checking the accompanying Lean certificates. It says all 38 headline theorems passed with no errors or non-standard axioms, while raising concerns about trusting definitions written by the same system that produced the proofs. ### Source excerpt OpenAI shipped ten open problems with Lean certificates. I built all 550,000 lines and checked what they rest on. Everything passed -- and the only thing left to trust is 1,700 lines of definitions written by the same system that wrote the proofs. ## Verifying a Claimed Counterexample to the Jacobian Conjecture DevFeed: [Verifying a Claimed Counterexample to the Jacobian Conjecture]() Original publisher: [Read original article]() Published: 2026-07-20T00:00:00Z Content type: opinion Language: en Sources: [Alex Korbonits]() Topics: [trust](), [Claude](), [anthropic]() Tags: [claude](), [model](), [trust](), [verify]() ### AI overview The author examines a claimed counterexample to the long-standing Jacobian conjecture, posted in a tweet and attributed to Anthropic's Claude model. They describe checking the polynomial map's constant Jacobian determinant and non-invertibility using symbolic calculations, while noting that no paper or referee report was available. ### Source excerpt The Jacobian conjecture fell yesterday, during the World Cup final, via a tweet. Wikipedia still says it's open. Here's how I checked it myself in two cells of sympy -- and why the fact that I could is the real story. ## Building a Minimal Control Plane to Reconcile Docker Containers for Model Serving DevFeed: [Building a Minimal Control Plane to Reconcile Docker Containers for Model Serving]() Original publisher: [Read original article]() Published: 2026-07-02T00:00:00Z Content type: tutorial Language: en Sources: [Alex Korbonits]() Topics: [control-plane](), [model-serving](), [Docker Container](), [Amazon Machine Learning](), [Streaming]() Tags: [container](), [control-plane](), [inference](), [model-serving](), [python]() ### AI overview The author explains control planes by building a small Python-based system that maintains a declared number of Docker containers for model serving. The system repeatedly compares desired and actual state and reconciles differences, including restoring a container after it stops. ### Source excerpt I can design batch, real-time, and streaming inference -- the data plane. But I'd never built the control plane that manages it. So I built the smallest one I could, from scratch, to finally understand reconciliation: the one idea that separates a control plane from a deploy script. ## AI proof systems can appear convincing despite mathematical errors DevFeed: [AI proof systems can appear convincing despite mathematical errors]() Original publisher: [Read original article]() Published: 2026-06-12T00:00:00Z Content type: opinion Language: en Sources: [Alex Korbonits]() Topics: [Artificial Intelligence](), [Large Language Model](), [math]() Tags: [ai](), [analysis](), [llm](), [math](), [model](), [verification]() ### AI overview The article examines MaxProof, a MiniMax system trained to generate competition-math proofs and improve them through candidate search and LLM-based verification. It reports a large gap between the training verifier's scores and independent expert judgments, while noting that the system's strongest contest results rely on extensive search rather than one-shot generation. ### Source excerpt Two posts ago I quoted a warning: an AI will find it easier to convince you it has a proof than to write one. A middling new paper finally put a number on that gap -- 0.99 against 0.55. ## Who Verifies the Verifier DevFeed: [Who Verifies the Verifier]() Original publisher: [Read original article]() Published: 2026-05-28T00:00:00Z Content type: opinion Language: en Sources: [Alex Korbonits]() Topics: [Artificial Intelligence](), [Formal verification](), [Compiler](), [Lean](), [Math and Logic](), [Google](), [Inference](), [Language models]() Tags: [ai](), [compiler](), [formal-verification](), [google](), [inference](), [model](), [paper](), [research]() ### AI overview The article examines whether formal verification can make AI-generated mathematical proofs scalable. It contrasts human review of natural-language proofs with Google DeepMind's approach of generating proofs directly in Lean and using the Lean compiler to verify them, while noting that the system's ability to read existing mathematics remains limited. ### Source excerpt An AI built the machine I said mathematics needed -- a compiler that verifies proofs for cents instead of expert weekends. The catch is what it still can't read. ## Same-Day Model Integration: ESMC + ESMFold2 in Sheaf v0.11 DevFeed: [Same-Day Model Integration: ESMC + ESMFold2 in Sheaf v0.11]() Original publisher: [Read original article]() Published: 2026-05-27T00:00:00Z Content type: article Language: en Sources: [Alex Korbonits]() Topics: [Large Language Model](), [structure](), [Pydantic](), [backends](), [unit tests](), [Serverless]() Tags: [backend](), [claude-code](), [code](), [language](), [model](), [pypi](), [release](), [serverless](), [structure](), [tests]() ### AI overview The article explains how Sheaf v0.11 integrated ESMC and ESMFold2 on the same day they were released. It attributes the rapid integration to existing typed serving contracts and infrastructure, while describing verification, testing, H100 smoke testing, release work, and documentation. ### Source excerpt Chan Zuckerberg Biohub released a new protein language model and structure predictor this morning. Sheaf v0.11 shipped with both, same day. The story is less about hustle and more about what a typed serving contract buys you when a new model lands. ## Verification Matters When AI Claims to Solve Mathematics Problems DevFeed: [Verification Matters When AI Claims to Solve Mathematics Problems]() Original publisher: [Read original article]() Published: 2026-05-23T00:00:00Z Content type: opinion Language: en Sources: [Alex Korbonits]() Topics: [Artificial Intelligence](), [math](), [OpenAI](), [Mathematics]() Tags: [ai](), [mathematics](), [openai](), [verification]() ### AI overview The article contrasts an unverified OpenAI claim about solving Erdős problems with a later result on the planar unit-distance conjecture that was accompanied by a paper and reviewed by mathematicians. It argues that verification, rather than simply generating proofs, is the central challenge as AI-produced mathematics becomes cheaper. ### Source excerpt An AI disproved one of Erdős's favorite conjectures. The interesting part isn't the proof -- it's who read it, and what happens when nobody can. ## Using Dolt's Versioned SQL Database for Feature-Store Point-in-Time Reads DevFeed: [Using Dolt's Versioned SQL Database for Feature-Store Point-in-Time Reads]() Original publisher: [Read original article]() Published: 2026-05-15T00:00:00Z Content type: opinion Language: en Sources: [Alex Korbonits]() Topics: [feature-store](), [Git](), [reproducibility](), [Databases](), [SQL]() Tags: [dedupe](), [feature-store](), [git](), [reproducibility](), [rfc](), [row-number](), [sql](), [sql-database]() ### AI overview This article describes a prototype Feast offline-store plugin backed by Dolt, a version-controlled SQL database. It argues that Dolt's revision-based reads can provide point-in-time feature retrieval and reproducible training snapshots without the usual append-only log deduplication CTEs. Tests on toy datasets produced identical results with shorter queries, though the article presents this as an early spike rather than a production benchmark. ### Source excerpt A weekend spike asked whether Dolt's AS OF reads could replace the ROW_NUMBER dedupe at the heart of every feature store's point-in-time join. Four weeks later, the RFC is quiet and the plugin's get_historical_features works end-to-end against a live Dolt server. Notes on building anyway. ## Planning an ML Survey Paper: Notes from a Notebook DevFeed: [Planning an ML Survey Paper: Notes from a Notebook]() Original publisher: [Read original article]() Published: 2026-04-30T00:00:00Z Content type: opinion Language: en Sources: [Alex Korbonits]() Topics: [Machine learning]() Tags: [notes](), [survey]() ### AI overview A candid account of planning an ML survey paper through three notebook sittings. The author describes choosing a topic, researching venues, drafting in Overleaf, missing an AMLC submission deadline, and abandoning the project after needing to find an internal transfer. ### Source excerpt An operational artifact, not a polished essay. The numbered checklist I worked out across three notebook entries in April 2024 -- what venue, what topic, how to find the bibliography, solo or co-author, milestones -- when I was deciding whether to write an ML survey paper. Useful for anyone in a similar spot. ## Three Things Made My Blog Agent-Ready. Five I Skipped on Purpose. DevFeed: [Three Things Made My Blog Agent-Ready. Five I Skipped on Purpose.]() Original publisher: [Read original article]() Published: 2026-04-30T00:00:00Z Content type: opinion Language: en Sources: [Alex Korbonits]() Topics: [Web](), [Netlify](), [Cloudflare](), [RSS Feed]() Tags: [ai-agents](), [blog](), [cloudflare](), [ietf](), [markdown](), [mcp-oauth](), [netlify](), [robots-txt](), [rss](), [web]() ### AI overview The author describes improving a personal blog's score on Cloudflare's agent-readiness checker from 25 to 50 by adding content signals in robots.txt and Link headers through Netlify. The article argues that personal blogs should implement applicable agent-facing features while skipping API, OAuth, and MCP requirements that do not fit their purpose. ### Source excerpt Cloudflare's isitagentready.com scored korbonits.com at 25 / Level 1 'Basic Web Presence.' Two hours and three small additions later it scored 50 / Level 4 'Agent-Integrated.' The other half of the points came from checks that don't apply to a personal blog -- and shipping fake compliance for them would be theatre, not value. ## The Geometry of Language: Embeddings as Manifolds, Writing as Geodesics DevFeed: [The Geometry of Language: Embeddings as Manifolds, Writing as Geodesics]() Original publisher: [Read original article]() Published: 2026-04-30T00:00:00Z Content type: article Language: en Sources: [Alex Korbonits]() Topics: [Embeddings](), [math](), [Large Language Model](), [Retrieval Augmented Generation (RAG)]() Tags: [amazon](), [embeddings](), [llm](), [math](), [rag]() ### AI overview The article explores a geometric view of writing improvement, treating strategic writing as a direction or submanifold in embedding space. It proposes a six-step experiment to measure strategic direction, find a geodesic path while preserving other properties, map the result back to tokens, and return the changes as guidance. ### Source excerpt What if 'good writing' is a direction in embedding space, and the model's job is to take the geodesic toward it? March 2024 notebook pages on Amazon Titan embeddings, question-manifolds, the curse of dimensionality, and a six-step experiment plan for treating strategic writing as a path-finding problem. ## Proofs and Essays Are Paths: An LLM ↔ Prover Loop for Falsifying Hallucinations DevFeed: [Proofs and Essays Are Paths: An LLM ↔ Prover Loop for Falsifying Hallucinations]() Original publisher: [Read original article]() Published: 2026-04-30T00:00:00Z Content type: opinion Language: en Sources: [Alex Korbonits]() Topics: [Large Language Model](), [Automated reasoning](), [Graphs](), [Artificial Intelligence](), [Math and Logic]() Tags: [grounding](), [hallucinations](), [knowledge-graph](), [llm](), [reasoning](), [smt](), [verification](), [verify]() ### AI overview This essay proposes a bidirectional loop between a large language model and a theorem prover. LLM outputs become propositions and graph connections that the prover checks, while verification results feed back into the model. In this framework, hallucinations appear as knowledge-graph paths with edges that fail formal verification. ### Source excerpt If a proof is a path through a formal space and an essay is a path through a semantic one, can we close the loop -- translate LLM outputs into a theorem prover, build a knowledge graph of verifiable propositions, and use the gaps as a signal for hallucination? Notebook ideas from March 2024 expanding on the SMT-grounding piece. ## Sheaf: vLLM for Non-Text Foundation Models DevFeed: [Sheaf: vLLM for Non-Text Foundation Models]() Original publisher: [Read original article]() Published: 2026-04-14T00:00:00Z Content type: article Language: en Sources: [Alex Korbonits]() Topics: [vllm](), [Inference](), [foundation-models](), [Large Language Model](), [Machine learning](), [model-deployment](), [Python](), [Streaming](), [Docker](), [lora](), [SIEM, Security, Observability]() Tags: [docker](), [foundation-models](), [gpu](), [inference](), [llms](), [lora](), [observability](), [python](), [streaming](), [vllm]() ### AI overview This article introduces Sheaf, a serving framework for non-text foundation models. It argues that vLLM's serving optimizations benefit from the shared compute pattern of autoregressive text LLMs, while time-series, tabular, molecular, biological, and diffusion models require different batching, memory-management, and inference approaches. The article describes Sheaf's proposed capabilities, including typed contracts, model-aware batching, streaming, caching, observability, offline batch inference, asynchronous workers, LoRA adapter multiplexing, a typed Python client, Docker and KubeRay deployment support, and 27 PyPI backends. ### Source excerpt vLLM solved inference for text LLMs. The same gap exists for every other class of foundation model -- time series, tabular, molecular, diffusion, and more. Sheaf fills it: typed contracts, model-type-aware batching, streaming, caching, observability, offline batch inference, an async-job worker, LoRA adapter multiplexing, a typed Python client, a Docker base image with KubeRay deployment, and 27 backends on PyPI. ## Using Claude Code to Add Ruff Linting and CI to 19 Open-Source Research Repositories DevFeed: [Using Claude Code to Add Ruff Linting and CI to 19 Open-Source Research Repositories]() Original publisher: [Read original article]() Published: 2026-04-12T00:00:00Z Content type: article Language: en Sources: [Alex Korbonits]() Topics: [Open Source](), [Claude Code](), [ci](), [pull-requests](), [Tooling](), [Development]() Tags: [ci](), [claude-code](), [code](), [dev](), [open-source](), [pull-requests](), [tooling]() ### AI overview The author describes using Claude Code to systematically add Ruff linting, pre-commit hooks, and CI workflows to 19 high-starred machine-learning research repositories. The article identifies recurring gaps between declared tooling and enforced checks, and explains how existing formatters, domain-specific annotations, and gitignored configuration files shaped the changes. ### Source excerpt How I used Claude Code to systematically add Ruff linting and CI to 19 high-starred research repos -- and what the week that followed revealed about the difference between correctness and value. ## The Philosophy of Machine Learning, or: What Comes After Hegel? DevFeed: [The Philosophy of Machine Learning, or: What Comes After Hegel?]() Original publisher: [Read original article]() Published: 2026-04-09T00:00:00Z Content type: opinion Language: en Sources: [Alex Korbonits]() Topics: [Automated reasoning](), [Artificial Intelligence](), [LLMs](), [Math and Logic](), [pattern matching]() Tags: [ai](), [automated-reasoning](), [conditional-probability](), [llms](), [logic](), [machine-learning](), [pattern-matching](), [reasoning]() ### AI overview This opinion essay compares large language models with Kant's distinction between a priori structures and a posteriori experience. It argues that LLMs learn patterns from human text but lack reliable logical scaffolding, and presents automated reasoning systems such as SMT solvers and theorem provers as a possible complement. ### Source excerpt A notebook entry that maps AI paradigms onto the history of Western philosophy -- from scholasticism to Hegel -- and asks what comes next when the current moment exhausts itself. ## Cosine Similarity vs Claude: What Happens When They Disagree DevFeed: [Cosine Similarity vs Claude: What Happens When They Disagree]() Original publisher: [Read original article]() Published: 2026-04-05T00:00:00Z Content type: comparison Language: en Sources: [Alex Korbonits]() Topics: [Embeddings](), [Claude](), [AI search]() Tags: [claude](), [embedding](), [search](), [semantic]() ### AI overview This article compares cosine similarity from embedding models with Claude's assessment of sentence meaning. Using "I love you" and "I hate you" as an example, it explains why embeddings can place antonyms close together and describes the distributional basis of modern sentence embeddings. ### Source excerpt I built semantic search for my blog and a dual-score similarity tool. The cases where a real embedding model and Claude disagree are more interesting than the cases where they agree. ## Using LLMs with SMT Solvers for Automated Reasoning and Formal Verification DevFeed: [Using LLMs with SMT Solvers for Automated Reasoning and Formal Verification]() Original publisher: [Read original article]() Published: 2026-04-02T00:00:00Z Content type: opinion Language: en Sources: [Alex Korbonits]() Topics: [Automated reasoning](), [Large Language Model](), [Formal verification](), [Prompt Engineering](), [Claude](), [graph theory](), [Machine learning](), [Artificial Intelligence]() Tags: [automated-reasoning](), [claude](), [formal-verification](), [graph-theory](), [llm](), [prompt-engineering]() ### AI overview The author revisits a November 2023 idea about using Claude and prompt engineering to translate problems into propositions for an SMT solver, satisfy proofs, eliminate unknowns and quantifiers, and return results to an agent. The essay connects this idea to the author's background in formal logic, graph theory and machine learning, while noting that the concept was not novel in isolation. ### Source excerpt A November 2023 notebook entry about grounding LLMs with SMT solvers -- why formal verification might be the right fix for hallucination, and what it took to finally say it out loud. ## I Deployed My First Smart Contract in an Afternoon DevFeed: [I Deployed My First Smart Contract in an Afternoon]() Original publisher: [Read original article]() Published: 2026-03-27T00:00:00Z Content type: tutorial Language: en Sources: [Alex Korbonits]() Topics: [Ethereum](), [smart contract](), [Solidity](), [Deployment](), [Tooling]() Tags: [deployment](), [ethereum](), [smart-contract](), [smart-contracts](), [software]() ### AI overview A first-principles walkthrough of building and deploying a minimal ERC-20 token to Ethereum, covering the token standard, balances, allowances, access control, tooling, and deployment requirements. ### Source excerpt A first-principles walkthrough of building and deploying a minimal ERC-20 token to Ethereum mainnet -- no frameworks, no abstractions, just Solidity, Hardhat, and Uniswap. ## Building an Open Source Video Generation Stack After Sora's Shutdown DevFeed: [Building an Open Source Video Generation Stack After Sora's Shutdown]() Original publisher: [Read original article]() Published: 2026-03-26T00:00:00Z Content type: opinion Language: en Sources: [Alex Korbonits]() Topics: [Open Source](), [Artificial Intelligence](), [OpenAI](), [Mixture of Experts (MoE)](), [GPU](), [Inference](), [Latency]() Tags: [ai](), [apache](), [architecture](), [benchmarks](), [generation](), [gpu](), [inference](), [latency](), [open-source](), [video]() ### AI overview The author argues that Sora's shutdown exposes the risks of relying on closed video-generation APIs and describes a planned open source replacement stack. The article compares Wan2.2, HunyuanVideo, and LTX-Video, discussing their architectures, quality, licensing, hardware requirements, and latency tradeoffs. ### Source excerpt OpenAI killed Sora with no warning and no migration path. Here's the open source video generation stack I'm building to replace it -- Wan2.2, LTX-Video, Ray Serve, and the embedding layer most writeups skip. ## Migrating from GitHub Pages to Netlify in an Evening DevFeed: [Migrating from GitHub Pages to Netlify in an Evening]() Original publisher: [Read original article]() Published: 2026-03-25T00:00:00Z Content type: tutorial Language: en Sources: [Alex Korbonits]() Topics: [migration](), [GitHub Pages](), [Netlify](), [GitHub](), [GitHub Actions](), [hosting](), [Astro]() Tags: [astro](), [dns](), [environment-variables](), [github](), [github-actions](), [github-pages](), [hosting](), [migration](), [netlify](), [secrets](), [ssl](), [workflow]() ### AI overview A first-person guide to migrating a static Astro blog from GitHub Pages to Netlify. It covers connecting the repository, transferring environment variables, moving DNS and SSL, restoring Mailgun email-forwarding records, and replacing a scheduled GitHub Actions rebuild with a Netlify build hook. ### Source excerpt Another hosting migration I'd been putting off for years, done in 30 minutes with Claude Code. ## Building a Self-Hosted Analytics Dashboard in One Session DevFeed: [Building a Self-Hosted Analytics Dashboard in One Session]() Original publisher: [Read original article]() Published: 2026-03-25T00:00:00Z Content type: tutorial Language: en Sources: [Alex Korbonits]() Topics: [dashboards](), [Self-hosted](), [Netlify](), [Supabase](), [Astro](), [Serverless](), [Dark Mode]() Tags: [analytics](), [building](), [cookies](), [dark-mode](), [dashboard](), [google-analytics](), [netlify](), [self-hosted]() ### AI overview A tutorial describing how to build a self-hosted analytics dashboard without Google Analytics or third-party tracking scripts. The implementation uses a beacon, a Netlify serverless function, Supabase, build-time rendering, and an authenticated analytics page. ### Source excerpt No Google Analytics, no third-party scripts, no cookies. Just a Netlify function, Supabase, and a Chart.js dashboard I actually own. ## I Shipped a Feature From My Phone During Lunch DevFeed: [I Shipped a Feature From My Phone During Lunch]() Original publisher: [Read original article]() Published: 2026-03-25T00:00:00Z Content type: opinion Language: en Sources: [Alex Korbonits]() Topics: [Claude Code](), [Vibe coding](), [JavaScript](), [Astro](), [GitHub Pages](), [API]() Tags: [api](), [astro](), [build](), [claude-code](), [development](), [github-pages](), [javascript](), [ml]() ### AI overview A developer describes building and deploying a stock price dashboard from a phone with Claude Code during a lunch break. The experience showed that AI can make implementation faster, while product judgment--such as choosing a useful data source--still requires human oversight. ### Source excerpt I shipped a working stock price dashboard from my phone during a lunch break using Claude Code -- and learned that execution is getting cheaper while judgment stays expensive. [Next page]()