DeepSeek open-sources DeepInfer, its V4.1-Flash serving stack

Daily AI Espresso gives preferred AI Weekly readers the signal from leading AI minds—without the noise.

Feedback is at the bottom. You can stop daily emails anytime.

DeepSeek has open-sourced DeepInfer, the framework it describes as "the high-performance inference framework that powers DeepSeek's large-scale production services," releasing it under Apache 2.0 on GitHub. The repo targets DeepSeek-V4.1-Flash and ships support for both Nvidia SM100 GPUs and Huawei Ascend 950 NPUs, the same Huawei silicon DeepSeek said last week it plans to pre-train on by end of 2026. The feature list reads as a bundle of DeepSeek's recent systems papers made operational.

Pick the topics, companies and people you track. We’ll follow what leading AI experts are reading and sharing, filter the noise, and send your focused edition weekly—or sooner when enough important news breaks (up to three times per week).

A developer typed a TV model number into Claude Code. The agent found the set on his Wi-Fi, removed 25 Xiaomi packages, and wrote a new launcher. That is the pitch for cleanyourtv, one of two GitHub projects TechCrunch reports consumers are using to strip ads, trackers, and bloatware from smart televisions. The other is dumbtv, built by Muzzammil Yasin Sabuwala using DeepSeek V4.1 Flash, Claude Opus 5.5, and Claude Sonnet 5.5, and tested end-to-end on a Fire TV Edition set and on Android/Google TV hardware from Sony, TCL, Philips, Hisense, Xiaomi, and Shield.

AWS's CFO joked about the financial toll of AI. An internal recording reveals why.

Stanford's Keshigeyan Chandrasegaran, Fei-Fei Li, Jiajun Wu and co-authors with Radical Numerics open Halcyon, a family of early-fusion multimodal autoregressive T2I models trained from scratch on GPIC, a 100M image-text dataset — no language pretraining. Halcyon-0.52B beats LlamaGen-XL/8 at 18.2x less pretraining compute (48.0 vs. 63.3 gFD); Halcyon-1.6B beats JiT-T2I-XXL/8++ at 275x less compute (35.3 vs. 41.0 gFD). Controlled experiments show text-only pretraining does not improve T2I generation.

article: Analysis arguing that artificial intelligence will never achieve consciousness due to fundamental limitations in cognitive architecture; cognitive science, consciousness, artificial intelligence theory.

A new foundation-model paper on Hugging Face claims it can cut the persistent storage needed to serve a 128K-token history by roughly 66x, by baking memory into the model itself rather than carrying a full KV cache. The paper, MemT-1.0: Towards Memory-Native Evolution, releases two models: MemT-1.0-Nano at 36B parameters with 3B active and 10M dynamic memory, and MemT-1.0-Flash at 288B parameters with 17B active and 336M dynamic memory. The backbones are Qwen3.6-35B-A3B and DeepSeek-V4-Flash respectively.

policy article: argues that governments could employ artificial intelligence to rebuild public trust, emphasizing citizen engagement; public policy, governance, trust, AI ethics

Meeting-notes tool Otter.ai raised $320M from General Catalyst's Customer Value Fund, a non-dilutive structure that returns capital based on marketing performance. The round dwarfs the roughly $50M Otter had raised across its decade of existence, with the company saying it crossed $100M ARR last year and grew to 40M users. Otter plans to pour the capital into enterprise go-to-market, international expansion, and new features like pre-meeting briefs, live AI assistance during calls, and workflow automation.

article: AI tool predicts individual cancer survival using a single patient selfie; medical AI, cancer prognosis, image-based prediction, healthcare analytics