Nat Taylor (@nattaylor) header
Nat Taylor (@nattaylor) profile picture

Product Manager, AI tinkerer, web developer, sailor, dad and CoCo'09 grad

Marblehead, MA
Joined October 2008
Photo by Nat Taylor (@nattaylor)Photo by Nat Taylor (@nattaylor)Photo by Nat Taylor (@nattaylor)Photo by Nat Taylor (@nattaylor)Photo by Nat Taylor (@nattaylor)Photo by Nat Taylor (@nattaylor)Photo by Nat Taylor (@nattaylor)Photo by Nat Taylor (@nattaylor)Photo by Nat Taylor (@nattaylor)Photo by Nat Taylor (@nattaylor)
baseten et al are already doing disagg + spec-dec + custom kernels, so what did they cook up? Better quantization? Scheduling?
we are popping the pricing bubble today starting with deepseek flash and glm 5.3 we are pricing 60% lower than fireworks/baseten while doing more volume than them all hosted on our own GPUs, there’s no reason to buy from anyone else we’re #1 now, good luck everyone else
20
Nat Taylor retweeted
We've seen decisions emerge as a really useful primitive in the ecosystem over the past few months for a wide range of use cases where full generation is overkill. That led us to build the Gemini Decisions API, bringing fast, calibrated decisions across text, images, audio, and video. Still early and iterating fast. Happy building and let us know what we should fix! aistudio.google.com/docs/dec…
Today we’re launching the Gemini Decisions API, powered by our new Gemini Decision 1 model in @GoogleAIStudio. Instead of generating text token by token, it’s built for closed-set choices scoring up to 1,000 questions across text, images, audio, video + PDFs in a single pass. A few details: - 3 decision types: yes/no, multiple choice (up to 255 options) + rubric scoring - $0.10 / 1M input tokens with $0 billed for output tokens (~5x cheaper than Flash-Lite) - leads on several benchmark evals with lowest calibration error Google teams are already building with it, below is a look at Stitch Rapid designing a full product flow in seconds (1x speed)
4
1
14
507
My AI agents found an unmapped Maya site in Mexico, using LiDAR data from a NASA dataset. 72 hours from a random thought on vacation to a reply from Mexico's national archaeology institute. What a time to be alive…
262
2,047
621
20,033
1,382,048
Nat Taylor retweeted
It’s not the kinda news starting with a J you’d like to hear, but... JEV models are now fully supported in transformers ! We standardized a bit so its should be simpler for everyone! See more details below
3
4
33
3,265
Last night my #Muse was prompt injected (non-maliciously) by a PDF Newsletter. The task was to parse the newsletter and extract articles, but one section was a series of questions. The response was answers to those questions, instead of verbatim text!
39
Nat Taylor retweeted
You can now generate, preview, and edit documents, PDFs, presentations, spreadsheets, and more directly in Gemini. blog.google/innovation-and-a… < you can say "ABOUT TIME" or "FINALLY" and that's ok. But I'm just happy it's now functionality we offer.
3
4
1
45
2,244
Nat Taylor retweeted
Meet NVIDIA Nemotron 3.5 Super VL Our new open model combines reasoning and tool use with text, image, document, and video understanding for long-running AI agents. 120B total parameters with 12B active 🧵
🤖 Made with AI
291
549
365
4,444
614,077
Should you waste tokens having your coding agent use grep, or hand the search to a tiny local model? We release ColGrep-Agent-2B! A code-search agent that runs locally, searches with ColGrep, reads files, & returns locations + code to your main agent -> 34% fewer API tokens
31
47
12
261
20,537
A 30M-parameter model, on CPUs, inside Postgres It matches a 0.6B dense embedding model on top-100 recall, to within 0.004, and finds more in the top 1,000 Meet Evoke. Open source today.
20
58
19
517
36,951
Nat Taylor retweeted
Introducing topk-embed-v1: a family of open-source and hosted text/image embedding models with frontier retrieval quality at $0.05/1M tokens. The models produce highly compressible representations, enabling us to serve them at scale with storage comparable to dense baselines.
10
27
5
184
19,329
Nat Taylor retweeted
CC the entire fam 🤝! Today, we’re announcing the new CC – an AI agent built for families to spend less time on logistics and more time together. You can now: 👤 Add up to 5 members to your CC agent ☀️ Start mornings aligned with a shared "Your Day Ahead" brief email 🗓️ Autosync schedules & to-dos with a shared Google Calendar and Tasks 💬 Coordinate in Google Chat with CC to offload relevant tasks (ie., crafting weekly meal plans, school supply shopping lists, etc) 📝 Delegate paperwork (ie., permission slips, forms, and more) for CC to complete under your direction 📌 Keep tabs on the details – CC remembers what applies to everyone (ie., family grocery lists, favorite restaurants) versus what applies to one person (ie., dietary restrictions, local timezones) Ready to keep everybody on the same page? Join the waitlist or upgrade your existing CC (US only, 18+): labs.google/cc
85
110
57
1,215
712,830
gpt-5.6-luna in a plant care guide: "General guidance, not a substitute for observing your plant." 🤣
42
6hr delay, board, deplane, board, taxi, return to gate and @AmericanAir has 2 bags of pretzels to offer. Maybe time to heed the engine's warnings and not try to push it trans-atlantic
1
77
Hope this isn't my last post. They gave it the ol' turn-it-off-turn-it-on once again and are trying again
1
39
The third time is a charm. We made it
13
Nat Taylor retweeted
Gemma 4 just got even faster! We're releasing Multi-Token Prediction (MTP) drafters that deliver up to a 3x speedup, without any degradation in output quality or reasoning logic.
95
335
64
3,282
208,659
New in Gemini: Generate files and export them Tell Gemini what you want to create and the format, and it now does the work for you. Now supporting: 📄 Google Docs, Word (.docx) & PDFs 📊 Google Sheets, Excel (.xlsx) & CSV 🖥️ Google Slides 🛠️ Markdown, LaTeX, TXT, RTF Available now on all surfaces globally!
151
274
63
2,920
340,621
Hi, we are releasing ColGrep 1.2.0 ColGrep now incorporate BM25 trigrams to further enhance our multi-vector models using hybrid search. Now, ColGrep print relative paths by default (fewer tokens per result) Exact same features as GREP Improved CUDA usage and installation
4
14
4
110
18,627
Amazing that reasoning can be compressed into the weights
💫 Introducing New SOTA Long Context VLM LightOn OriOn-Qwen-SR1 reasons over full documents and executes it implicitly at inference. Reasoning is compressed into the model's weights, no verbose output, no added latency. 🥇 SOTA on MMLongBenchDoc, ahead of Qwen3 VL with 7× fewer parameters. 🙌 Kudos to @further_ai for this new milestone! Reasoning starts at reading. 👉 lighton.ai/lighton-blogs/doc…
1
49
Nat Taylor retweeted
This April Fool's Day, we decided to stop joking around. Trinity-Large-Thinking is out now.
20
36
10
374
70,016