Helping organize @AIconference #raysummit #NLPsummit #DataAISummit ; Past: Co-Chair of @strataconf @OReillyAI @TensorFlowWorld & #SparkAISummit
San Francisco, CA
Joined December 2008
- Tweets14.3K
- Following383
- Followers41.2K
- Likes234
Pinned Tweet
🆕 What Teams Building AI Agents Keep Learning the Hard Way
🎯 the emerging playbook for production AI Agents
gradientflow.com/i-keep-hear…
🎙️ Akshay Sharma and Nick Ung @lyft share how they built a platform that lets domain experts (not just engineers) create customer-facing AI agents. The key? Guardrails, prompt linting, and a robust eval pipeline.
thedataexchange.media/akshay…
The Rise of Deep Model Customization
🔎 The most interesting shift in these projects: teams treat the base model as replaceable. Thomson Reuters expects to start from newer open models as they appear. C3 AI has repeated its process across five generations. Cursor reused its training machinery for Composer 2.5. The durable asset is the model factory, not any single model.
gradientflow.substack.com/p/…
Interesting WSJ short documentary on political shifts in Greenwich, CT
youtu.be/hW5UjNHCma0?is=2UOc…
Enterprise AI and Deep Model Customization.
🔎 Companies including Thomson Reuters, Cursor, C3 AI, Cisco, and Reddit are developing AI models that understand their specific data, workflows, and products.
👉 gradientflow.substack.com/p/…
Deep model customization is emerging as a real pattern. Companies like Thomson Reuters, Cursor, C3 AI, ByteDance, Cisco, and Reddit are all going well beyond basic fine-tuning.
👉 gradientflow.substack.com/p/…
Amir Bar-Lev delivers another standout doc with The Last First. This one is on complex human dynamics, hubris, and commercial exploitation on K2. I wasn't aware of how a Nepalese team outmaneuvered ("deceived") & struck out on their own, to summit first
youtube.com/watch?v=zgGpyr5M…
👍👍 Tenzing a new 🎬 by Jennifer Peedom and Luke Davies
🎥 this is a moving portrait of the man who reached Everest’s summit alongside Edmund Hillary, and follows the true story of their historic 1953 climb
en.wikipedia.org/wiki/Tenzin…
Running AI on your own devices or internal servers is becoming more realistic
🎯 Local Models Are Improving. That Doesn't Mean They Go Everywhere.
When you run a model locally, proprietary code and internal business data don't leave your infrastructure. That's a meaningful shift for companies that handle sensitive information and want to know exactly where it's being processed.
→ gradientflow.com/the-local-a…
an audience question from Rachel Reeves to Paul Krugman
youtube.com/live/yYITJnqWkz0…
Will AI wipe out jobs in a sudden wave, or is the reality far more nuanced? @MIT FutureTech director @ProfNeilT shares ground-level empirical data on how AI adoption actually unfolds across occupations.
thedataexchange.media/neil-t…
Local AI Is Improving. Where Should Businesses Run It?
gradientflow.com/the-local-a…
How is AI really impacting the workforce? @ProfNeilT @MIT explains why AI's effect on most jobs will be a "rising tide" rather than a "crashing wave."
#AI #FutureOfWork #MIT #AIEconomics #BusinessTech #aitraining #Careers
thedataexchange.media/neil-t…
Been testing Gemini 4 Argon over the last few days 🔎
Verdict: It’s surprisingly good and definitely worth including in your mix of proprietary frontier models.
That said, long-term? Still firmly Team Open-Weights 🙌
Before You Move AI to the Edge, Read This
🔎 Where should each AI task run: on a laptop, an internal server, or with an outside provider? Narrow, easy-to-check work such as transcription and document classification may fit employee devices.
→ gradientflow.substack.com/p/…
Running AI Closer to Home
🎯 Local AI isn't about replacing the cloud. It's about matching each task to the right place: laptop, internal server, or outside provider. Getting that routing right matters more than the hardware specs.
→ gradientflow.substack.com/p/…
