Sam Keen (@samkeen) profile picture

Software Engineer, AI researcher/writer, @Databricks, opinions my own.

Portland, OR, USA
Joined May 2007
Sam Keen retweeted
A year ago, I ran @databricks totally differently from today. Getting stuff done at Databricks involved calling ppl, scheduling meetings, and lots of back and forth. Today, I just wake up and use Genie, which works for me in the background. I often just use the phone app to queue up work in the background. Three examples from just this week: → Org charts. How do we flatten the org? It literally found me the hotspots where many projects had gotten stuck. Turns out the data was already in the ontology graph. → New internal project everyone raves about (Keanu). Genie monitored the slack channel, got all the insights, correlated them with the source code, and gave me a dashboard of what I needed. → New culture survey. Used ontology to figure out from the org chart, work output, and survey scores, which leaders are AI pilled. Why certain scores (e.g. work-life balance) are so low. These are things that used to take me hours and many meetings, often waiting for the data science team. This is possible because Genie has access to an Agentic Ontology, so it proactively does all the work in the background for me. Learn more about our next-generation release of Genie: databricks.com/blog/new-geni…
16
85
10
377
37,057
Sam Keen retweeted
𝗜𝗻𝘁𝗿𝗼𝗱𝘂𝗰𝗶𝗻𝗴 𝘁𝗵𝗲 𝗻𝗲𝘄 𝗚𝗲𝗻𝗶𝗲: an always-on, autonomous AI coworker that proactively handles long-running work within your organization's guardrails. AI assistants can answer questions and run analyses. But most enterprise work is complex and continuous, requiring teams to track changes, follow up, and take action. Genie Ontology is now generally available, providing the shared business context behind it. The agentic ontology now includes new knowledge sources and new ways to curate and govern that context. New capabilities across Genie Agents and plugins extend that same context to more workflows and AI tools. Here's what's new: databricks.com/blog/new-geni…
40
40
11
208
18,979
Sam Keen retweeted
Kimi K3.1 Pro is available on Databricks on day 0. Build governed AI apps and agents on Lakehouse data with Kimi K3.1 Pro and 70+ other models. Unity Gateway lets you govern, monitor and secure calls, and test models without changing app code.
Releasing the model weights of Kimi K3.1 Pro: huggingface.co/moonshotai/Ki… We'll keep advancing open intelligence.
6
12
5
65
17,710
Trying some AI local dev Flint is burning lots of reasoning tokens but it's accomplishing it's goals and none of this is going to @AnthropicAI 's reported 85% margins
1
19
Bit of a quiet News week in AI last week, but still some great content in the Tutorials and Tools space. writing.alteredcraft.com/p/w… This week's stories are making up for it. Subscribe to Altered Craft to get a human curated weekly review every Monday
21
This week's edition of Altered Craft's weekly review is out.
20
Nice to see the inclusivity Both slow and fast Zombies welcome eventbrite.com/e/zombie-scav…
28
Interesting split happening in AI models: alongside the big reasoning models, there's a growing class of small ones that just decide things (Jev's). Microsoft-Decision-1 does routing, classification, prioritization and verification, returning structured output software can act on. Microsoft reports it at 2.5x the speed of the runner-up in their benchmarks. commandline.microsoft.com/mi…
14
Wikimedia's bandwidth use is up 50% since 2024 because of bot traffic, and 65% of its most resource-heavy requests now come from bots. Undisclosed bot edits in sandbox areas, edits to a citation tool's config to misuse it as a proxy, failed attempts to abuse a public Etherpad, and millions of API calls that may have contributed to a Wikidata Query Service outage in May. The cost of agents behaving unpredictably lands on volunteers and non-profits, not the companies shipping them. I'm not against AI, but please do better Frontier Labs diff.wikimedia.org/2026/10/0…
1
22
Claude Code (the app) is going through some major overhauls... You can now write small JS modules that live inside a Claude Code session and see every event as it happens. They work like middleware. Your hook runs, then other plugins, then Claude Code itself. So you can watch a tool call, rewrite it, or refuse it. This guide builds a context window forecast drawn above the prompt in about 80 lines. claude.dev/blog/getting-star… #ClaudeCode #HarnessEngineering
1
47
Several leading closed models score near zero on a security benchmark because they refuse the task, not because they can't do it. @MistralAI's launch post for Mistral Large 4, a 1 trillion parameter multimodal model with 49 billion active parameters, weights due by the end of the month. On a test that asks a model to reproduce a real vulnerability in open-source software and then patch it, ML4 scores 82%. Defenders often need to prove a flaw is real before they can fix it, and provider-level refusals block that work. Open weights plus self-deployment means you set the policy. mistral.ai/news/mistral-larg… #AI #OpenWeights
10
OpenJev (running locally) vs Typesafe Jev in a dungeon crawl text adventure test harness writing.alteredcraft.com/p/t…
42
Join us at the next Cascadia Builders Club meetup We are a community for people building hardware and software. This is show-and-tell for builders: short, hands-on demos of what you're actually making (apps, hardware, robotics, XR), and then we hang out and meet the people behind the work. No pitch, no hype reels, just real things and the people making them. cascadiabuilders.club/ #Portland #Robotics #Meetups
21
A speech recognition model that fits in 16.9 MB, runs on plain CPU, and reaches its first token in 11 milliseconds. Introduced Whistle on the @cactuscompute blog. It handles seven languages, word timestamps and speech embeddings. cactuscompute.com/blog/whist…
2
27
Benchmarks against Whisper base and Moonshine tiny are in the post, with the gaps shown in both directions. In the same spirit, I also covered a new STT model from Microsoft AI in the last Weekly review: writing.alteredcraft.com/i/2…
16
Nice, in the revamped @claudeai projects a subagent can start a subagent. It’s subagents all the way down! #ClaudeCode #Agents
1
22
And yet the Decision Model mania continues. Now with a frontier lab entering the ring. @OpenAI admitted that currently the Decisions API is backed by a optimized Luna and harness and was in reaction to @typesafeai 's Jev
1
33
That said, there were some other very interesting events in AI last week. Check out Altered Craft's weekly review, a human curated list the the top AI Tutorials, Tools, and News of the previous week. writing.alteredcraft.com/p/w…
6