Publication

Towards Deep Learning

@towards-deep-learning

articles: 48
total claps: 9,365
top claps: 2,053
avg read: 10.1 min
paywalled: 33
active: 2026-04-15 → 2026-07-16

Wiki topics

AI · AI · General · 40 ML · Machine Learning · 35 EDU · Education & Learning · 32 LLM · Large Language Models · 27 🔬 · Science · General · 18 AGT · AI Agents · 15 💻 · Programming · 13 🔓 · Open Source · 8

Top tags

#artificial-intelligence ·38 #deep-learning ·26 #ai ·22 #machine-learning ·21 #data-science ·18 #llm ·14 #large-language-models ·7 #ai-agent ·5 #claude-code ·4 #generative-ai-tools ·4

AI TCH ECO

SU Sumit Pandey · Towards Deep Learning

RIP Commercial OCR. An Open-Source Model Just Topped Every Benchmark.

A 4B parameter model from a Brooklyn startup is beating GPT-4o and Gemini across 90 languages. Although OCR is not my domain, this one…

Apr 15, 2026 · 6 min read · 2,053 claps paywalled #ocr #large-language-models #llm

RIP Commercial OCR. An Open-Source Model Just Topped Every Benchmark.

AI HUM MDA BIO TCH

SU Sumit Pandey · Towards Deep Learning

A Single CLAUDE.md File Went Viral. The Reason Is Embarrassingly Simple.

91,000 stars on GitHub. No code. Four rules from Andrej Karpathy that every coding agent should have been following from day one.

May 08, 2026 · 4 min read · 1,930 claps paywalled #artificial-intelligence #deep-learning #ai

A Single CLAUDE.md File Went Viral. The Reason Is Embarrassingly Simple.

AI SPT TCH

SU Sumit Pandey · Towards Deep Learning

Why I Stopped Using Gemma 4 and Switched to Qwen 3.6

Why Qwen’s new Mixture of Experts model is the first open-source LLM that actually handles agentic coding workflows

Apr 25, 2026 · 8 min read · 815 claps paywalled #gemma-4 #qwen #large-language-models

Why I Stopped Using Gemma 4 and Switched to Qwen 3.6

AI HUM

SU Sumit Pandey · Towards Deep Learning

NVIDIA’s 3B Model Just Embarrassed 30B Vision Models

It detects anything you can describe, runs 10x faster than its rivals, and fits in a third of the size. The trick: it stopped spelling out…

May 29, 2026 · 11 min read · 517 claps paywalled #artificial-intelligence #deep-learning #machine-learning

NVIDIA’s 3B Model Just Embarrassed 30B Vision Models

AI HUM TCH

SU Sumit Pandey · Towards Deep Learning

DSpark: DeepSeek Made LLMs Faster Without Changing a Word

DSpark, DeepSeek’s open-source speculative decoding framework, makes V4 inference up to 85% faster with no quality loss.

Jul 01, 2026 · 12 min read · 386 claps paywalled #large-language-models #llm #artificial-intelligence

DSpark: DeepSeek Made LLMs Faster Without Changing a Word

AI HUM ART TCH

SU Sumit Pandey · Towards Deep Learning

Moebius: The Image Inpainter Model That Beats 50x Of Its Size

Moebius is a 226M open-source inpainting model that rivals the 11.9B FLUX.1-Fill-Dev and runs over 15x faster. Here is how.

Jul 02, 2026 · 8 min read · 370 claps paywalled #deep-learning #artificial-intelligence #machine-learning

Moebius: The Image Inpainter Model That Beats 50x Of Its Size

AI SCI HUM ART TCH SPT

SU Sumit Pandey · Towards Deep Learning

PewDiePie Released His “Trillion-Dollar” AI Project for Free

PewDiePie’s Odysseus is a free, open-source AI workspace running better than ChatGPT-style tools on your own PC. Here’s what it does.

Jun 03, 2026 · 9 min read · 302 claps paywalled #ai #artificial-intelligence #chatgpt

PewDiePie Released His “Trillion-Dollar” AI Project for Free

AI TCH

SU Sumit Pandey · Towards Deep Learning

Claude Code Without Subscription: A Proxy That Actually Works

A small open-source proxy points Claude Code at NVIDIA NIM, Kimi, GLM, MiniMax, and friends. Zero subscription. Two-minute setup. Honest…

May 19, 2026 · 14 min read · 273 claps paywalled #claude-code #generative-ai-tools #free-ai-tools

Claude Code Without Subscription: A Proxy That Actually Works

AI HUM SCI

SU Sumit Pandey · Towards Deep Learning

The One Trick That Made My Code Bug-Free: Step by Step Explanation

How I turn a battle-tested CLAUDE.md into a CLAUDE.md + flow.md, and one review pass into Claude Code sessions that ship almost no bugs

Jun 30, 2026 · 10 min read · 255 claps paywalled #ai-agent #artificial-intelligence #llm

The One Trick That Made My Code Bug-Free: Step by Step Explanation

AI HUM TCH SCI

SU Sumit Pandey · Towards Deep Learning

I Gave Fable 5 and GPT-5.6 the Same Build Task, Results Were Shocking

Claude Code worked on its own and finished quickly, but codex needed more push and it build a more advanced, higher-quality product.

Jul 14, 2026 · 8 min read · 249 claps paywalled #vibe-coding #artificial-intelligence #data-science

I Gave Fable 5 and GPT-5.6 the Same Build Task, Results Were Shocking

AI SPT HUM SCI

SU Sumit Pandey · Towards Deep Learning

Gemma 4 12 b: Google Released The Model Without Encoder And That was My WTF Moment

Gemma 4 12B is an open, encoder-free multimodal model that runs on a 16GB laptop. Here is what that actually means, and why it matters.

Jun 08, 2026 · 12 min read · 246 claps paywalled #artificial-intelligence #ai #data-science

Gemma 4 12 b: Google Released The Model Without Encoder And That was My WTF Moment

AI TCH HUM

SU Sumit Pandey · Towards Deep Learning

Microsoft Open-Sourced Something It Wasn’t Supposed To, Then Deleted Everything

The strange case of FastContext, the Microsoft coding-agent tool that cut tokens by 60% before it quietly vanished online.

Jul 06, 2026 · 9 min read · 229 claps paywalled #artificial-intelligence #machine-learning #ai

Microsoft Open-Sourced Something It Wasn’t Supposed To, Then Deleted Everything

AI HUM MDA SCI

SU Sumit Pandey · Towards Deep Learning

Oxford Researchers Showed AI Hallucinations Are Predictable, Here is How ?

This paper reframes LLM hallucinations as compression failures, then predicts and gates them with an information budget.

Jun 26, 2026 · 11 min read · 207 claps paywalled #data-science #artificial-intelligence #llm

Oxford Researchers Showed AI Hallucinations Are Predictable, Here is How ?

AI HUM SCI

SU Sumit Pandey · Towards Deep Learning

Claude Fable 5: Anthropic Released Its Most Powerful and Feared Model

Anthropic just shipped Claude Fable 5 and Mythos 5: the same brain wearing two faces, one safe for everyone and one locked away.

Jun 09, 2026 · 10 min read · 161 claps paywalled #llm #artificial-intelligence #ai

Claude Fable 5: Anthropic Released Its Most Powerful and Feared Model

AI SCI HUM

SU Sumit Pandey · Towards Deep Learning

NVIDIA Just Gave Robots a Brain That Can See the Future

Cosmos 3 reasons about a scene, dreams up what happens next, and decides how to move. One model does all three.

Jun 01, 2026 · 10 min read · 156 claps paywalled #artificial-intelligence #deep-learning #machine-learning

NVIDIA Just Gave Robots a Brain That Can See the Future

AI SCI

SU Sumit Pandey · Towards Deep Learning

GPT-5.6 Just Shocked the Math World Wrote a Proof of a 50-Year Old Math Problem in Under an Hour.

OpenAI says its newest model proved a famous 50-year graph theory conjecture in under an hour. No one has verified it yet.

Jul 12, 2026 · 12 min read · 141 claps paywalled

GPT-5.6 Just Shocked the Math World Wrote a Proof of a 50-Year Old Math Problem in Under an Hour.

AI HUM MED SCI

SU Sumit Pandey · Towards Deep Learning

How Microsoft’s AI Replaces a $5,000 Cancer Test With a $5 Slide

A new model turns ordinary pathology slides into virtual immune maps, making cancer research cheaper, faster, and much bigger.

Jun 03, 2026 · 8 min read · 135 claps paywalled #artificial-intelligence #machine-learning #ai

How Microsoft’s AI Replaces a $5,000 Cancer Test With a $5 Slide

AI HUM SPT SCI

SU Sumit Pandey · Towards Deep Learning

Nemotron 3 Ultra: Nvidia’s New AI Loses Half Its Benchmarks But Wins Anyway, Here is Why ?

Nemotron 3 Ultra is Nvidia’s open 550B agent model: 5x faster and up to 30% cheaper on long-running AI agent tasks.

Jun 04, 2026 · 12 min read · 108 claps paywalled #artificial-intelligence #llm #ai

Nemotron 3 Ultra: Nvidia’s New AI Loses Half Its Benchmarks But Wins Anyway, Here is Why ?

AI HUM

SU Sumit Pandey · Towards Deep Learning

Google Just Changed Everything for Machine Learning on Spreadsheets

TabFM is Google’s zero-shot foundation model for tabular data: it predicts on unseen tables in a single pass, no training.

Jul 05, 2026 · 11 min read · 95 claps paywalled #artificial-intelligence #ai #machine-learning

Google Just Changed Everything for Machine Learning on Spreadsheets

AI TCH

SU Sumit Pandey · Towards Deep Learning

Harness Engineering: The Quiet Coding Skill Everyone Will Need

The change happened without me noticing, and I think it matters more than the next model.

May 22, 2026 · 6 min read · 94 claps paywalled #ai-agent #coding #llm

Harness Engineering: The Quiet Coding Skill Everyone Will Need

AI HUM TCH SCI

SU Sumit Pandey · Towards Deep Learning

Tired of Claude Code’s Over-Engineering ? Ponytail is Here For You

A free skill for Claude Code and other AI agents that fights code bloat. An honest look at whether the 54% claim holds up.

Jun 25, 2026 · 10 min read · 93 claps paywalled #coding #deep-learning #artificial-intelligence

Tired of Claude Code’s Over-Engineering ? Ponytail is Here For You

AI HUM

SU Sumit Pandey · Towards Deep Learning

RLMs: The MIT Trick That Makes a Small AI Beat GPT-5

A plain-English guide to Recursive Language Models (RLMs): what they are, how they work, and why giving an LLM a Python environment beats…

May 26, 2026 · 12 min read · 80 claps paywalled #large-language-models #artificial-intelligence #machine-learning

RLMs: The MIT Trick That Makes a Small AI Beat GPT-5

AI HUM

EL Elsie Rainee · Towards Deep Learning

I Used Agentic AI vs Generative AI: Here’s the Difference

Generative AI creates. Agentic AI executes. The difference matters more than you think.

Jul 15, 2026 · 7 min read · 78 claps #ai-agent #ai #artificial-intelligence

I Used Agentic AI vs Generative AI: Here’s the Difference

AI TCH SCI

SU Sumit Pandey · Towards Deep Learning

I Built a Monster CLAUDE.md, And My Coding Agent Got Scary Good

How a gstack CLAUDE.md, built on Karpathy’s coding rules and harness engineering, helped me beat a published algorithm.

Jun 16, 2026 · 9 min read · 70 claps paywalled #claude-code #programming #coding

I Built a Monster CLAUDE.md, And My Coding Agent Got Scary Good

AI ECO

SU Sumit Pandey · Towards Deep Learning

Andrej Karpathy: Man Who Taught a Generation to Build Neural Nets Just Joined Anthropic

The OpenAI co-founder could have joined any lab on Earth. The team he’s building tells you why this one matters.

May 20, 2026 · 4 min read · 55 claps paywalled #ai-news #andrej-karpathy #anthropic-claude