<![CDATA[Stories by Sumit Pandey on Medium]]> https://medium.com/@sumit.ai?source=rss-73d7984df4b2------2 https://cdn-images-1.medium.com/fit/c/150/150/1*66cJkvGe-pOi8_tQ2f5FKg.jpeg Stories by Sumit Pandey on Medium https://medium.com/@sumit.ai?source=rss-73d7984df4b2------2 Medium Sun, 15 Nov 2026 02:35:50 GMT <![CDATA[I Tested SenseNova U1.5 Against ChatGPT. It Failed Just One Image Test]]>

I ran SenseNova locally on a DGX Spark against ChatGPT. It nailed text, handled complex edits, but couldn't count coffee cups.

Continue reading on Towards Deep Learning »

]]> https://www.towardsdeeplearning.com/i-tested-sensenova-u1-5-against-chatgpt-it-failed-just-one-image-test-b3720b6635cd?source=rss-73d7984df4b2------2 https://medium.com/p/b3720b6635cd Fri, 13 Nov 2026 14:12:09 GMT 2026-11-13T14:12:09.580Z <![CDATA[AI on Excel Sheet: NVIDIA’s Kumo Runs 23x Faster Than Google’s TabFM]]>

It predicts straight from your spreadsheet with no training at all, and unlike the two rivals that edge past it, it’s free for business…

Continue reading on Towards Deep Learning »

]]> https://www.towardsdeeplearning.com/ai-on-excel-sheet-nvidias-kumo-runs-23x-faster-than-google-s-tabfm-21c0d20518ee?source=rss-73d7984df4b2------2 https://medium.com/p/21c0d20518ee Sun, 08 Nov 2026 19:35:52 GMT 2026-11-08T19:35:52.338Z <![CDATA[How uzu Runs a 9B Model 4x Faster Than llama.cpp.]]>

Memory bandwidth caps a base M5 at about 29 tokens a second, so here’s how uzu gets 92 and where the 4x needs an asterisk.

Continue reading on Towards Deep Learning »

]]> https://www.towardsdeeplearning.com/how-uzu-runs-a-9b-model-4x-faster-than-llama-cpp-f429bfd493bb?source=rss-73d7984df4b2------2 https://medium.com/p/f429bfd493bb Fri, 06 Nov 2026 13:05:53 GMT 2026-11-06T13:05:53.316Z <![CDATA[Weight-Space Programming: How GPT-6 Astra Fixed Qwen by Editing 0.016% of Its Weights]]>

MIT researchers handed OpenAI’s GPT-6 Astra the raw weights of a small Qwen model and asked it to make the model better at math, and it…

Continue reading on Towards Deep Learning »

]]> https://www.towardsdeeplearning.com/weight-space-programming-how-gpt-6-astra-fixed-qwen-by-editing-0-016-of-its-weights-1727e0666824?source=rss-73d7984df4b2------2 https://medium.com/p/1727e0666824 Wed, 04 Nov 2026 19:56:01 GMT 2026-11-04T19:56:01.680Z <![CDATA[Andrej Karpathy Got Sick of YouTube Slop, So He Built a HackerTube]]>

HackerTube is 6,454 videos that Hacker News actually recommended, and Karpathy didn’t write, read or run a single line of the code behind…

Continue reading on Towards Deep Learning »

]]> https://www.towardsdeeplearning.com/andrej-karpathy-got-sick-of-youtube-slop-so-he-built-a-hackertube-88b0c0e190d5?source=rss-73d7984df4b2------2 https://medium.com/p/88b0c0e190d5 Tue, 03 Nov 2026 10:02:07 GMT 2026-11-03T10:02:07.636Z <![CDATA[OpenShell: Nvidia's Open-Source Answer to Rogue AI Agents at OpenAI and Anthropic]]>

After rogue agents hit Hugging Face and Medicare, Nvidia’s open-source OpenShell runtime puts the fence outside the agent

Continue reading on Towards Deep Learning »

]]> https://www.towardsdeeplearning.com/openshell-nvidias-open-source-answer-to-rogue-ai-agents-at-openai-and-anthropic-7aa19a8e8817?source=rss-73d7984df4b2------2 https://medium.com/p/7aa19a8e8817 Wed, 21 Oct 2026 23:46:42 GMT 2026-10-21T23:46:42.321Z <![CDATA[My Supervisor Posted a Meme About Our Quantization Paper. Here’s the Story]]>

How a postdoc experiment inspired by Google’s TurboQuant grew into ApexQuant, a quantizer that compresses AI models without looking at a…

Continue reading on Towards Deep Learning »

]]> https://www.towardsdeeplearning.com/my-supervisor-posted-a-meme-about-our-quantization-paper-heres-the-story-6c67d2d75e9d?source=rss-73d7984df4b2------2 https://medium.com/p/6c67d2d75e9d Tue, 13 Oct 2026 07:54:10 GMT 2026-10-13T07:54:10.770Z <![CDATA[Qwen-Image vs SenseNova U 1.5: I Gave Two Best Image Models 5 Tasks.]]>

Continue reading on Towards Deep Learning »

]]> https://www.towardsdeeplearning.com/qwen-image-vs-sensenova-u-1-5-i-gave-two-best-image-models-5-tasks-e96b9af193c5?source=rss-73d7984df4b2------2 https://medium.com/p/e96b9af193c5 Fri, 09 Oct 2026 09:12:33 GMT 2026-10-11T13:43:59.808Z <![CDATA[Opus 5.5 vs GPT-6: I Made Opus 5.5 and GPT-6 Grade Each Other’s Work]]>

Claude Opus 5.5 vs GPT-6-Astra on coding, creative writing and a game build: who won and the bugs they found in each other

Continue reading on Towards Deep Learning »

]]> https://www.towardsdeeplearning.com/opus-5-5-vs-gpt-6-i-made-opus-5-5-and-gpt-6-grade-each-others-work-779c1cc9fb57?source=rss-73d7984df4b2------2 https://medium.com/p/779c1cc9fb57 Tue, 06 Oct 2026 18:11:48 GMT 2026-10-06T18:11:48.159Z <![CDATA[OpenAI Just Turned on Its Partners, and Even Its Own Former Researcher]]>

OpenAI’s DevDay 2026 launches, from Dots and Pages to the Decisions API and Codex code review, now compete with Notion, CodeRabbit, Jev…

Continue reading on Towards Deep Learning »

]]> https://www.towardsdeeplearning.com/openai-just-turned-on-its-partners-and-even-its-own-former-researcher-2ccd2338681e?source=rss-73d7984df4b2------2 https://medium.com/p/2ccd2338681e Fri, 02 Oct 2026 12:31:05 GMT 2026-10-02T12:31:05.560Z