SnackOnAI Blog

Aug 30, 2026

OpenSEO Is Very Good At Three Percent Of AI Citations

AI engines cite peer and competitor pages twenty six times more often than they cite your own domain. Every SEO tool you own works on the domain.

Aug 29, 2026

Anthropic's Alignment Agents Did Not Out-Think Anyone

A full automated research run costs about what one human idea costs, and produces a hundred and seventy five of them. That is the whole result.

Aug 26, 2026

Qwen3-VL And The Multimodal Tax

An eight billion parameter model with tools beat the flagship without them. The report has the numbers and does not connect them.

Aug 25, 2026

Rex-Omni, Or What Reinforcement Learning Actually Fixed

A day of GRPO on eight GPUs delivered the headline gain. It improved tight box precision by one tenth of a point.

Aug 24, 2026

FreeToken, Or Why Your Laptop Was Never The Bottleneck

An eight gigabyte laptop GPU just decoded a 35B MoE model at ninety two percent of an RTX 4090's rate, and the reason has nothing to do with the GPU.

Aug 11, 2026

LinkedIn Benchmarked Nine Memory Systems For Its Hiring Agent, Seven Lost To Pasting Everything Into The Prompt

LinkedIn published a table this spring that quietly indicts most of the agent memory market. Their own paper, their own benchmark, their own baselines. Almost nobody has read the rows that matter.

Aug 9, 2026

Autograd Is Slow Per Operation And Free Per FLOP, And That One Sentence Tells You When To Use It

The reverse-mode engine that taught the field how to build autodiff is under two hundred lines of Python. You can read all of it in an afternoon. What you cannot do is make it fast, and the reason is architectural, not a missing optimization.

Aug 8, 2026

The Seedance Paper Is a Scoreboard, Not a Blueprint

ByteDance shipped a twenty six page paper on Seedance 2.0 containing twenty three pages of benchmark tables and zero equations.

Aug 7, 2026

AnyDoc Is Not Faster Than Docling. It Is Solving A Different Problem, And Almost Nobody Noticed

anydoc beats Docling by a factor of one hundred on document conversion speed. That number is real, reproducible in spirit, and almost completely misleading.

Aug 6, 2026

LFM2: Liquid AI's Hardware-in-the-Loop Architecture Search Found That SSMs Were Not the Answer for Edge Devices, and the Result Runs 2x Faster Than Qwen3 on a Galaxy S25

Most efficient model architectures are designed on paper, benchmarked on A100 clusters, and then discovered to behave differently on the actual target hardware.

Aug 1, 2026

HeadRoom: The Context Compression Layer That Sits Between Your Agent and Your LLM Bill, and Gets 92% Token Reduction Without Touching Your Code

Every coding agent you run is sending far more tokens to the LLM than the task requires. A code search returning 100 results generates 17,765 tokens. The LLM needs 1,408 of them to find the answer.

Jul 31, 2026

OpenWorker: Andrew Ng Built the AI Coworker That Delivers Finished Files, Not Chat, and Put the Agent Loop on Your Machine

Every AI assistant tool available today delivers the same output: a text response. You paste it somewhere, format it, send it, file it, or act on it yourself.

Jul 30, 2026

CAMEL: The First Multi-Agent Framework That Used Its Own Agents to Generate the Data That Trained Its Competitors

The dominant narrative around multi-agent frameworks is that you pick one based on features: does it support tool calling, memory, RAG, streaming?

Jul 29, 2026

Colibri: A Single C File Runs a 744B MoE on 25GB of RAM by Treating Every Expert as Data to Be Staged, Not State to Be Held

The assumption every MoE inference system makes is wrong. It assumes the model must fit in memory.

Jul 28, 2026

SkillOpt: Microsoft Trained a Markdown File to 80.7 on SpreadsheetBench, Up From 41.8, Without Touching a Single Model Weight

The dominant assumption in agent improvement is that better agents require better models. More parameters, more RLHF, more fine-tuning. SkillOpt (microsoft/SkillOpt, MIT, 13k stars, arXiv:2605.23904) challenges this from the opposite direction: freeze the model entirely, and train the skill document the same way you would train a neural network, with rollout batches, minibatch reflection, bounded updates, a held-out validation gate, and epoch-wise regularization.

Jul 27, 2026

Attention Residuals: Moonshot AI Fixed the Residual Connection That Every Deep LLM Has Been Getting Wrong Since 2017

Every transformer you have ever used accumulates layer outputs the same way: add them all together with fixed weight 1. Layer 3 contributes exactly as much as layer 47. Layer 1's token embedding contributes the same as the layer right before the output.

Jul 26, 2026

LLM Observability Is Not a Dashboard Problem. It Is a Five-Layer Integration Problem That Nobody Has Solved Yet.

You can monitor a web service with four metrics: request latency, error rate, CPU utilization, and memory usage. Those four numbers tell you almost everything you need to know. An LLM in production breaks all four of those assumptions simultaneously. A model can produce fluent, syntactically correct output that is factually wrong.

Jul 25, 2026

SIE: Superlinked's Inference Engine Solves the Wrong Problem That Every Other Serving Stack Was Built For

vLLM, SGLang, and TGI are built for one large model spread across many GPUs. Agents need the opposite: many small models sharing one GPU, switching on demand with sub-second cold start.

Jul 24, 2026

DevOps Open Agent: The AI Troubleshooter That Refuses To Trust Its Own AI

Most "AI-powered" DevOps tools fail because they trust the LLM too much. This one is interesting because it treats the LLM as a hostile witness.

Jul 23, 2026

OpenShip: The Self-Hosted Deployment Platform That Builds Locally and Ships Containers, Leaving Your Servers Free to Do One Job

Your production server runs Coolify. Coolify runs your apps. Also Coolify runs the CI. Also Coolify runs the dashboard. Also Coolify runs the build agent and the queue and the metrics collector.

Jul 22, 2026

Paperclip: The Agent Orchestration Layer That Solves the Problem Nobody Talks About, Which Is That Nobody Talked to the Agents Before Sending Them to Work

The tagline is exact: "If OpenClaw is an employee, Paperclip is the company."

Jul 20, 2026

Kimi K3 and Mooncake: Moonshot AI Shipped the World's First Open 3T-Class Model on a KVCache-Centric Inference Engine That Gets 525% More Throughput by Treating Cache as the Primary Citizen

Every LLM serving paper optimizes for throughput. Mooncake optimizes for cache. The distinction sounds subtle. It is not. When you make KVCache (key-value cache, the memory structure that stores intermediate attention computations) the first-class citizen of your serving architecture, you stop thinking about GPU clusters as compute nodes and start thinking about them as a heterogeneous memory hierarchy.

Jul 19, 2026

ModelExpress: NVIDIA Dynamo's Rust-Based Weight Management Layer Transfers a 70B Model Between GPUs Faster Than Loading It From Disk. The JIT Cache Transfer Is the Feature Nobody Is Talking About.

What used to be a minutes-long startup problem becomes a RDMA transfer measured in seconds.

Jul 17, 2026

Inkling: Thinking Machines Lab Built a 975B MoE With Controllable Thinking Effort, Relative Position Embeddings, and Short Convolutions on the Residual Stream. The Self-Fine-Tuning Demo Is the Real Signal.

Inkling (thinkingmachines/Inkling, open-weights, July 15, 2026) is Thinking Machines Lab's first model release: a 975B-total/41B-active Mixture-of-Experts transformer with a 1M token context window, encoder-free multimodal inputs (audio as dMel spectrograms, vision as 40x40 pixel patches via 4-layer hMLP), controllable thinking effort (a float you pass at inference time), and 30M+ RL rollouts shaping its behavior.

Jul 16, 2026

OpenScience: The Open-Source AI Workbench Launched Five Days After Claude Science. It Supports More Models, More Skills, and Runs on Your Infrastructure. The Tradeoff Is Everything That Comes With Being Five Days Old.

OpenScience (synthetic-sciences/openscience, Apache 2.0, v1.2.5, YC W26, openscience.sh) is a model-agnostic AI workbench for scientific research that runs the full research loop: literature review, hypothesis, code, experiment, analysis, and write-up, in one continuous session. It ships 250+ editable skills across ML, computational biology, cheminformatics, and cloud compute, plus 30+ scientific databases (UniProt, PDB, ChEMBL, arXiv, OpenAlex, Semantic Scholar) as native agent tools. Any frontier or open-weight model works with a single configuration flag; switching is per-request.

Mohinish S

Serverless Ventures | Cloud, Data & Distributed Systems | Angel & Advisor | Infra & Data Startups