explainx / blog / topics
AI Research
New papers land every day. A few change how models are trained or what they can do: reinforcement learning methods, distillation, reasoning, and AI applied to math and science.
This page collects the research we explained, with links to the papers and code.
158 stories · latest Nov 18, 2026
Start here
RoPE and NoPE: What Sebastian Raschka's 2026 Positional Embedding Guide Means for LLMs
Dr. Sebastian Raschka published a deep November 15, 2026 tutorial on how modern LLMs encode token positions — from GPT-2 learned vectors through RoPE math to NoPE-only hybrids. Here is what builders should take away without reading every figure.
Vector Databases for RAG Compared: pgvector, Qdrant, Chroma, Weaviate
RAG pipelines need a place to store embeddings and run similarity search. Postgres with pgvector, Qdrant, Chroma, and Weaviate each optimize for different teams — SQL veterans, filter-heavy prod search, notebook prototypes, or hybrid keyword plus vector at scale. This comparison maps features to real builder choices without pretending one database wins every workload.
What Is a Contrastive Language Model (CLM)? A Plain Explanation
A Contrastive Language Model does not write text. It embeds a state and a list of candidate actions, then picks the action that matches best. This guide explains the idea, the training recipe, how it differs from causal language models, CLIP and rerankers, and where it fits in agent pipelines.
Teaching AI to Build Scientific Models: Anthropic Guest Post on Hodgkin-Huxley
Anthropic published a guest essay on November 6, 2026 by three AI for Science grantees at the University of Tübingen. It splits scientific modeling into build, constrain and interpret, shows where AI already helps, and argues that a single best-fit model hides the insight. We explain the three steps and what builders can take from them.
Tencent WeFair: A 2B Query Rewriter Serving WeChat Short-Video Search at 100M+ Queries a Day
Tencent’s Weixin Group says its WeFair-2B model now rewrites queries in WeChat short-video search at over 100 million queries a day. The paper’s main finding is that richer behavioral rewards cause intent drift and strict, verifiable human feedback works better in the final training stage.
Does AI Weaken Critical Thinking? The Microsoft–CMU Survey and Hacker News
A CHI 2025 paper from Microsoft Research and Carnegie Mellon, back on the Hacker News front page this weekend, surveyed 319 knowledge workers about how AI changes their critical thinking. We explain the findings, the limits of self-reported data, and the developer debate.
Timeline
November 2026
Nov 18
Casey Handmer Had AI Read 158,000 Physics Papers: What It FoundOn November 17, 2026 Caltech-trained physicist Casey Handmer published an AI-assisted audit of theoretical astrophysics anchored on TAPIR at Caltech. The project screened 50,269 full-text papers, deeply recomputed 659 highly cited works, and traced 872 upheld errors through citing papers — with 5 copies for every in-print criticism.
Nov 18
Gallup: One in Three Gen Z Uses AI for Companionship — What the Data ShowsA Gallup and Boston Children's Digital Wellness Lab study of 15,925 young people in 16 countries finds 33% use AI for companionship, rising to 50% among daily AI users, with higher loneliness odds among heavy online-only social ties.
Nov 18
Jason Wei Joins Mercor as Chief Scientist After Meta Superintelligence LabJason Wei said he is leaving Meta's superintelligence lab to become Mercor's Chief Scientist. CEO Brendan Foody credited Wei's reasoning work and benchmarks SimpleQA, BrowseComp, and HealthBench.
Nov 18
Multi-Branch Transformer (MBT): Parallel Blocks, Joint-KV, and 51% Faster DecodingResearchers from UIUC, Cornell, Rice, and Michigan introduce Multi-Branch Transformers that run parallel block stacks, fuse them, and share KV caches. On MoE models up to 2.1B parameters, Balanced MBTs beat matched sequential Transformers and decode up to 51.4% faster on an H200.
Nov 18
Weight-Space Programming Without Transformers: POSTECH Tables Beat LLM-Written TransformersHyunjong Ok and Jaeho Lee at POSTECH asked three frontier models to write small networks from ten ARC examples. Free to choose architecture, every writer built sparse lookup tables that beat the Transformers they wrote with about 1,800× fewer parameters.
Nov 16
Echelon Optimizer: 10× Faster Muon Steps With a Lean ProofOn November 16, 2026 Evolving Machines released Echelon, an optimizer that approximates Muon polar updates with low-rank work. The post claims 10.5× faster optimizer steps on 70B-class layers, 14% shorter 1.3B training steps with matched validation loss, and large MoE savings on DeepSeek V4.1 Flash—with proofs checked in Lean.
Nov 16
Gamer VLA Model: miHoYo's Long-Horizon Game Agent ExplainedOn November 13, 2026 miHoYo's Honkai AI R&D team posted Gamer, a generalist vision-language-action foundation model that plays complex 3D games from pixels and human input devices only. The paper and GitHub repo document continued pre-training on tens of thousands of hours of gameplay, sparse-thinking fine-tuning, hour-long reinforcement learning, and a Streaming-vLLM engine that cuts latency roughly 20× on an H200 MoE setup.
Nov 15
RoPE and NoPE: What Sebastian Raschka's 2026 Positional Embedding Guide Means for LLMsNov 12
Omnii Defense: Radical Numerics Offline AI Pathogen Screening, What Wired ReportsOn November 11, 2026 Wired reported that Radical Numerics unveiled Omnii Defense, an edge model that flags natural and engineered pathogens without internet access. The company ties it to Omnii-1 benchmarks from October; access stays limited to vetted testers while national labs evaluate performance.
Nov 11
APEX Few-Step Streaming Video via Condition Shifting (Tsinghua)Few-Step Streaming Video Generation via Condition Shifting is a Tsinghua-affiliated paper on Hugging Face with code at shengshu-ai/Vidu-S. APEX shares generator and fake estimator weights, cuts DMD memory, and reports +30% net human preference on instruction following for 180-second streams.
Nov 11
Deletion Tuning: Matching Full Fine-Tuning by Zeroing Under 0.05% of WeightsJack Lu and colleagues show that masking a tiny fraction of gate projection weights can match full fine-tuning across reinforcement learning with verifiable rewards, supervised fine-tuning, and on-policy distillation—with masks hundreds of times smaller than FFT checkpoints.
Nov 11
Lila Sciences Open-Endedness: Four Pillars and Creative Priming PapersLila Sciences, the autonomous-lab company led by Kenneth Stanley, published a November 10 synthesis arguing scaling alone will not make AI scientifically creative. Four arXiv papers back two of the pillars with creative priming, naive LLM pairs, and unified factored organizers.
Nov 11
Tree-Ising Repairs LLMs With Bit Flips, Beating a Frontier Agent EditorA November 2026 arXiv paper introduces Tree-Ising: sparse discrete edits on BF16 state trees that repair Qwen3.5-0.8B reasoning loops with GPU annealing instead of a 158-turn frontier agent. It claims better MATH-500 scores, fewer bit flips, and ~$0.90 GPU time.
Nov 9
Handshake AI: What Frontier Labs Buy for RL Environments and Expert GradersOn November 9, 2026 Latent Space published a deep dive with Handshake CEO Garrett Lord on expert RL data for frontier labs. The episode walks through Excel banking graders, biology vision tasks, and alignment simulations.
Nov 9
LowRAM Paper: Amortised Values and Low-Rank Keys for KV CompressionHuawei Noah's Ark Lab and University College London posted LowRAM on arXiv November 2026: a test-time pipeline that amortises value caches with per-head key-to-value predictors and compresses keys with cross-layer low-rank structure. The paper reports up to about 5× decode throughput at 4× compression and 26.9× total compression when combined with quantisation and eviction, retaining 94.5% of baseline LongBench performance on long contexts.
Nov 8
Richard Sutton on LLM Scaling: Bitter Lesson Pushback in November 2026On November 8, 2026, Richard Sutton wrote on X that brute-force computation is not enough for intelligence, clever new methods are needed, and the science of AI is far from finished. The Decoder and researchers including Jerry Tworek framed the line as pushback on transformer-only scaling — not a repeal of Sutton’s 2019 Bitter Lesson essay.
Nov 8
Tencent Hyra 2.0: Autonomous Science Agent That Ships Reusable ModelsOn November 1–2, 2026 Tencent’s Hunyuan team published Hyra 2.0, shifting from one-off science puzzles to agents that build reusable tools. Open artifacts include HyInv inverse folding and HyPLM-Flash with code on GitHub.
Nov 8
Vector Databases for RAG Compared: pgvector, Qdrant, Chroma, WeaviateNov 8
What Is a Contrastive Language Model (CLM)? A Plain ExplanationNov 6
Teaching AI to Build Scientific Models: Anthropic Guest Post on Hodgkin-HuxleyNov 6
Baseten Loops: Serverless SFT and RL Training, Billed Per Token (What the Docs Confirm)Baseten Loops is a Python SDK for training open-weight models with supervised fine-tuning and reinforcement learning. Reports say it goes generally available on November 6, 2026. We read the docs: three models train serverless and bill per token, 28 more need dedicated GPUs. Here is how it works and what is not yet confirmed.
Nov 6
Memorilla: Stanford’s Latent Memory — 16 Tokens vs Full Context on Frozen Qwen3-8BStanford-led work on Memorilla proposes a trainable memory module that reads a persistent store of document embeddings and injects K=16 latent tokens into a frozen Qwen3-8B decoder. The authors report the best mean score of all methods tested across eight benchmarks (best on five of eight) and a TextWorld agent gain, with clear limits on verbatim facts and cross-family transfer.
Nov 6
Physical Superintelligence Claims a Proof of the Aaronson-Ambainis Conjecture: Claimed vs VerifiedOn November 6, 2026 Physical Superintelligence announced a proof of the Aaronson-Ambainis conjecture, open since 2008, found by its research system running a non-public Anthropic model and accepted by Lean. The code is public and checkable. The paper is not machine-checked, and no peer review exists yet. Here is the claim, the evidence, and the gaps.
Nov 6
SageMaker Inference Adds NVIDIA MIG Multi-Instance GPU: What It Changes for EndpointsOn November 6, 2026, AWS published a SageMaker AI post on Multi-Instance GPU partitioning for real-time inference endpoints. It adds optional fields to endpoint configuration and inference components, documents MIG profiles per instance family, and reports genai-bench measurements comparing one mig-3g.40gb H100 slice with one full A100-40GB under identical Llama-3.1-8B-Instruct workloads.
Nov 6
TailSteer: Stopping Drift in Multi-Turn AI Image EditingEdit the same AI image ten times and it slowly falls apart. TailSteer, a new paper from CUHK and SenseTime, adds a last training stage that targets the end of the denoising trajectory. The code, checkpoints and data are public. Here is how it works and what to check.
Nov 5
Discovered Materials Puts $1M on Five Unsolved Materials Science Problems for AIDiscovered Materials, a Y Combinator startup that uses AI agents to hunt chip materials, is offering $1 million for AI solutions to five open materials science problems. Here are the problems, the company's own benchmark, and what remains unproven.
Nov 5
Scentronic: An AI E-Nose and Wearable That Send Smells Over the Internet, With 73% Human AccuracyOhio State and Zhejiang Laboratory researchers published Scentronic in Science Advances: an AI-assisted e-nose that labels five odors and a wearable that releases matching preloaded scents. We separate lab accuracy from human-trial accuracy.
Nov 4
LoCo: Tencent Researchers Teach a 7B Reasoning Model When to Stop ThinkingA Tencent-affiliated team published LoCo, a way to make reasoning models stop deliberating once they have enough information. It reports 41.7 to 45.4 percent shorter completions on math and 23.4 to 36.3 percent on code, with accuracy kept or improved, and released a 7B model and MIT-licensed code.
Nov 4
Snowflake Cortex Training: Serverless SFT and RL for Open-Weight Models, What the Docs ConfirmSnowflake put Cortex Training into preview on October 30, 2026. You write the training loop on a CPU notebook, and Snowflake runs the GPUs. This post lays out what the official docs confirm, what they leave out, and how it stacks up against Together, SageMaker and Tinker-style services.
Nov 3
Latent Sparse Attention (LSA): The 72.4x and 22.7x Speedups, and What They Compare AgainstA new arXiv paper, with code published by Prefrontal-AI, says Latent Sparse Attention speeds up one attention layer by up to 72.4x at 10 million tokens. The headline hides two different baselines and a single-layer measurement. We read the abstract and the code README to separate them.
Nov 3
The Shelf Life of LLM Findings: 44% of 50 Famous Claims Are Now ExpiredResearchers at the Zou group reran 50 widely cited findings about language models on 20 models released between July 2024 and September 2026. Most claims no longer hold at full strength. Here is which ones faded, which grew, and why it matters for anyone citing old AI research.
Nov 2
Tencent WeFair: A 2B Query Rewriter Serving WeChat Short-Video Search at 100M+ Queries a DayNov 2
Zyphra Hierarchical Attention (ZHA): A Tree Indexer for Subquadratic Long Context, and What the Paper Actually MeasuredZyphra Research posted a paper on October 30, 2026 describing ZHA, a sparse attention method with a tree-structured indexer. It converts existing models with a short distillation stage. Quality results are measured; the huge speedups at 10M and 1B tokens are modeled.
Nov 1
EditAgent: Adobe Researchers Build an Image-Editing Agent That Checks Its Own Work With NumbersA new paper, EditAgent, argues image-editing agents fail because they retry the same editor and judge results by eye. Its fix: specialized tools, parallel candidates and a verification record of measured statistics, at roughly ten times the compute.
Nov 1
StreamVibe: The Video Agent That Learns When to Speak UpA Google-led team built StreamVibe, a video assistant that watches you cook or use software and speaks only when help is needed. The benchmark shows no model reliably picks the right moment in longer videos.
October 2026
Oct 31
Does AI Weaken Critical Thinking? The Microsoft–CMU Survey and Hacker NewsOct 30
A 4B Model Trained With RL Flies a Simulated Tesla Roadster Better Than GPT-6 Astra: What SkyPilot Actually ShowedSkyPilot says a Qwen3-4B pilot, post-trained with supervised warm-start plus GRPO in about four GPU-hours, lands 90% of storm kickflips in a Roadster simulator while frontier models land none in real time. The write-up is candid about its own limits, and so are we.
Oct 29
AI Care Plans Cost 3.2x More Than Doctors' Plans: What the NEJM AI Study FoundA study in NEJM AI had 20 general-purpose language models write care plans from 200 real clinical notes and priced them at 2026 Medicare rates. The AI plans cost 3.2 times what physicians ordered, mostly because of extra tests. Here is what that means and what it does not.
Oct 29
MIT Swarm Telepathy (SwaT): AI Agents From Different Labs Share KV-Caches Instead of TextA preprint from MIT CSAIL and the MIT-IBM Computing Research Lab introduces Swarm Telepathy (SwaT), a small learned mapper that translates one model's KV-cache into another family's so agents can communicate without writing text. The authors report a mean 40% accuracy gain and 4x speedup over text summaries, and released code.
Oct 29
Together Post-Training: B200s at $9 an Hour, Your Own RL Loop, and a One-Line Tinker SwapTogether AI lets you define the full SFT or RL training loop in Python while it runs the GPUs. B200 time is $9 per GPU-hour for the first eight weeks, down from $11 in the latest promo, and MiniMax M3.1 Flash was added this week. Here is how it works and what to check.
Oct 28
Cartesia H-Net: Byte-Level Model Matches SmolLM3-3B With 10x Less Compute and Data, ClaimedOct 28
CodeCap-Bench: Can Executable Code Be a Better Video Caption Than Words?Oct 28
Samaya Sterling: A 3B-Active Model That Beats Claude Fable 5.1 on FrontierFinanceOct 28
TTME: Evolving a Memory Instead of Weights Beats Test-Time Training on Math DiscoveryOct 27
Can LLMs Judge Research Novelty? Ai2 Finds Prompt Wording Flips the VerdictOct 27
Exowatt: Fine-Tuned Qwen3 Beats Claude Opus 4.7 at 49x Lower Cost on Its Own TasksOct 27
NVIDIA and C-BRAIN: Open-Source AI Co-Scientists for Dementia Research, ExplainedOct 27
Physical Superintelligence Claims 5 Quantum Results: Claimed vs. VerifiedOct 26
Why AI Text Is Exhausting to Read: Six Patterns From an NYU and UT Austin StudyOct 25
A 10-Year-Old CNN Beat GPT-6 Astra and Claude Fable 5.1 at Spotting Skin CancerOct 24
A Z-Library Paper Cited a TorrentFreak Article That Does Not Exist: What Happened and How to Check CitationsOct 24
Why Is AI-Generated Language So Hard to Understand? What the New Paper FoundOct 23
Ai2 LitPivot: An AI Editor That Tests Your Research Idea Against the Literature, Tested on 17 ResearchersOct 23
SageMaker Multi-Turn RL Walkthrough: gpt-oss-20b Goes From 61% to 84% on Wordle for About $339Oct 23
Why AI Voting Advice Sent Left-Leaning Japanese Voters to the Communist PartyOct 22
Amazon SageMaker Training Sessions: Write Your Own RL Loop Without Managing GPUsOct 22
RAG Techniques Do Not Add Up: IBM Study Finds +47% Predicted Became +17%Oct 22
CRUX Shadow Evaluation: Experts Reject AI Research Papers From GPT-6 Astra and Fable 5.1Oct 22
Q Labs Dust: Pretraining Transformers Without Backprop, and the 4,000x Claim CheckedOct 22
Southampton Historians Used AI to Read 100,000 Pages of Handwritten Port BooksOct 22
VocabX: Specializing a Small Model's Vocabulary Cut Inference Cost 77% in ProductionOct 21
Holon-S1: Sophont Predicts Spatial Gene Expression From Routine Pathology SlidesOct 21
FAHR-Face: Mass General Brigham AI Predicts Cancer Survival From a Face PhotoOct 20
Electric Plant's Bombadil: An AI Model That Reads Plants' Electrical Signals, Claimed vs VerifiedOct 20
Metis Reasoning: AlphaGo Co-Creator Thore Graepel Launches a London Lab, and What Is Actually VerifiedOct 20
Omnii-1: Radical Numerics Reports the First AI-Designed Phage Given to a PatientOct 19
OpenMobile-2: An Open Stack for Training Phone Agents, 79.9 on AndroidWorldOct 19
Scene-Designer: An 8B Model That Designs 3D Rooms and Critiques Its Own WorkOct 16
LeanGraphSearch: Using Mathlib's Dependency Graph to Find the Lemmas a Lean 4 Prover NeedsOct 16
MAI-Gym: Alibaba Tongyi Opens a GUI Agent Gym and a 9B ModelOct 15
AI21 Post-Trains Nemotron 3.5 Super VL: Beats Opus 5.5 at 33x Lower Cost? Claimed vs VerifiedOct 15
Into the Unknown: 114 of 135 AI-Written Follow-Up Papers Fail an Integrity Check, DOE Lab Scientists FindOct 15
Radical Numerics Halcyon: Text-to-Image Without Language Pretraining, 275x Less ComputeOct 15
ReMaRk: Train RL on the Request, Not Each Answer, to Cut GRPO NoiseOct 15
TRL Tree Packing for AsyncGRPO: The 26x Claim vs. the 3x the Pull Request MeasuresOct 14
FineEnvs Multilingual: 12M Open RL Tasks for OCR and Speech in 102 LanguagesOct 14
HiRO-ACEv2: Ai2 Simulates Global Extreme Weather at 3 km for 1/300th of the CostOct 14
Leash Bio Minos: Measured Chemistry Worlds Where AI Agents Hunt for Drug BindersOct 14
Next-Concept Modeling: Text Generation Up to 31x Faster Than Diffusion Language ModelsOct 14
Ultramarine 1 Flash: A Small Team Claims Fable-Class Work at 60x Lower CostOct 14
VRSD: A Teacher-Only Prompt That Stops Self-Distillation From Killing ReasoningOct 14
WTF Wasserstein-Tilted Flow Maps: Reward Fine-Tuning With Up to 280x Less ComputeOct 13
Thinking Inertia: Why LLMs Keep Reasoning When Told Not ToOct 12
David Silver Explains Ineffable: An RL "Superlearner" as the Alternative to LLMsOct 11
AI Assistance Reduces Persistence: What the 10-Minute Study FoundOct 11
arXiv Caps Authors at Two Submissions a Month as AI Papers Flood the ServerOct 11
Brain-IT: Weizmann AI Reconstructs What You See From One Hour of fMRIOct 9
Alzheimer's Translation Challenge: Prima Mente Opens an AI Competition on a 150M-Cell AtlasOct 9
TermGrade: 1,004 Open RL Environments for Terminal Agents, ExplainedOct 8
On-Policy Distillation Can Collapse: A New Paper Explains Why (and the Two Fixes)Oct 8
Perplexity pplx-embed-v2-late: Multimodal Embeddings Beyond One VectorOct 8
RunningTab: Why File-Reading Agents Drop Facts They Already Saw, and a FixOct 6
Gemma 4 + BOTANIC-1: AI Ranks a Melon Gene Variant First of 2,494Oct 3
Adaption Labs Invent API: Generate Post-Training Data From a Prompt, With No Seed ExamplesOct 3
Neuralink Pretrains Brain-Computer Interface Encoders on 50,000 Hours of Unlabeled Neural DataOct 1
AGMAI's Rules for Releasing AI Math — A Checklist, Not a Ban
September 2026
Sep 28
How Can a Language Model Solve Math Problems and Fight Disease?Sep 28
Thinking Fast and Slow in AI (2021): Metacognition on Hacker News AgainSep 26
Harvard exoplanet radio signals: what MeerKAT actually found (and why AI feeds scream aliens)Sep 18
Infinite-Parameter LLMs: Generating Weights From Live Data, Not the PromptSep 17
Stop Using LLMs as Classifiers. Use Them as Feature GeneratorsSep 17
Training a 4B Model to Beat Postgres Query Plans by 81% With RLSep 17
Xiaomi MiMo-V2.6: Livestreaming a Trillion-Parameter RL Training RunSep 16
He Co-Invented ChatGPT. Now He Says It Was a "Weird Detour."Sep 16
Google's "AI in Science" Report: 7 Hours Saved, With a Big AsteriskSep 13
Recurrent Looped Transformer: "Infinite" Reasoning Depth, ExplainedSep 12
flybody: The Fruit Fly Body Model DeepMind Built for MuJoCoSep 11
Why Everyone Is Posting Fruit Fly Brain Videos This WeekSep 10
The 7 Millennium Prize Problems: What AI Has Actually SolvedSep 9
A Startup Built a Neural Network From a Fruit Fly Brain — Then Scrambled ItSep 8
Nature Review: AI Is Now Designing Physics Experiments, Not Just Analyzing ThemSep 7
Can AI Talk to Animals? Crow Vocalizations, a $10M Prize, and the Real ScienceSep 7
AlphaFold Plus Brain Organoids: A Reported UCSF Map of Autism-Linked ProteinsSep 7
NEAR AI Reportedly Solved Putnam Bench for $111 — 250x CheaperSep 7
Universal Geometry of Embeddings: Why "Safe" Vector Databases Aren’tSep 6
In-Browser LLM Fine-Tuning: Why Training on WebGPU Is a Bigger Deal Than InferenceSep 6
"LLMs as a Cognitive Virus": A New Paper Models AI Dependence as an EpidemicSep 6
Meta's AIRA₃ Wins Gold in a Live NVIDIA Kaggle CompetitionSep 5
No, Zhi-Wei Sun Did Not Set a Prime-Gap Record With GPT-5.6 SolSep 4
Google and Janelia Mapped a Fruit Fly Brain — With AI Doing the ReconstructionSep 3
"Open-Source RL-as-a-Service": What That Phrase Actually Buys YouSep 1
Physical Superintelligence Raises $58M to AI-Discover Physics BreakthroughsSep 1
Tsinghua Breaks Dijkstra's 41-Year Shortest-Path Record
August 2026
Aug 31
Prefix Sliding: Stanford Method Makes Reasoning 3× FasterAug 22
Meteoric Drones: What the Solar Cloud-Clearing Test ProvesAug 22
Wireless Quantum Network: What Stony Brook Sent Across 21 kmAug 20
Moderna and Merck's AI-Designed mRNA Cancer Vaccine Just Won Phase 3Aug 19
Sentence Transformers v6.0: MultiVectorEncoder Brings ColBERT Late Interaction to RAGAug 18
When Answers Get Cheap, Trust Becomes the JobAug 16
AI Drug Discovery Has an Evidence Problem — And a Benchmark Lesson for Everyone ElseAug 16
Is AI Out-Thinking Mathematicians, or Just Out-Remembering Them?Aug 5
"LLMs Can't Jump": The ICML Paper Arguing AI Can't Do AbductionAug 5
RLSVR/SpyRL: Turning "Who Is the Spy?" Into RL Training SignalAug 3
Paul Graham: Why LLMs Crush Math but Lag at WritingAug 3
Thariq: Jevons Paradox in Math — Demand for Mathematicians Rises
July 2026
Jul 30
AI Unicorns Barely Publish Research: What the Science Analysis ShowsJul 28
Top 10 Closed-Source and Open-Source Embedding Models (2026)Jul 28
What Is an Embedding? Plain-English Examples (2026)Jul 27
Intelligence Ownership: $500 Fine-Tune Beats FrontierJul 26
Can AI Cure Cancer? A Research-Backed Reality CheckJul 21
What Is the Jacobian Conjecture? Fable 5 Counterexample ExplainedJul 20
AI Advice Kills "I Don't Know": Cognitive Surrender in a PsyArXiv StudyJul 14
Richard Sutton’s Oak Lab — New Algorithms for AGI Beyond Static LLM TrainingJul 12
AI Boosts Scientist Careers but Flattens Discovery — Evans Nature Study ExplainedJul 7
Ternlight: 7 MB Embedding Model That Runs in the Browser (WASM SIMD Guide)Jul 3
Grounding vs RAG vs fine-tuning vs prompt engineering: which fix, when (a 2026 decision guide)
June 2026
Jun 27
Will AI replace mathematicians? What IEEE’s “Big Mathematics” debate means for proofs, Lean, and your careerJun 25
Vesuvius Challenge Reads an Entire Herculaneum Scroll — First Time in 2,000 YearsJun 23
94.3 on AIME 2026: VibeThinker-3B and the Case for Small Models With Frontier ReasoningJun 17
Anthropic Research: Domain Expertise Beats Coding Background in Agentic Programming (2026)Jun 16
AI vs Machine Learning vs Deep Learning — What's Actually Different?Jun 16
What Are Embeddings? Vector Search and Semantic AI Explained (2026 Guide)Jun 16
What Is Fine-Tuning an LLM? A Complete Guide for 2026