Freedom Dumlao

CTO + CAIO. Ruby Central Board Member. Maker. Builder. Opinions my own.

[Freedom Dumlao @APIguy]

How many additional tokens will I have to pay for with these new "watermarks"? 🤔

[Freedom Dumlao @APIguy]

You may have noticed Ive been quiet here lately... Heres why!

The team and I have been heads down re-imagining the future of how wealth managers work. We are so, so excited to share this with everyone!

[Vestmark @Vestmark]

Meet Vestmark Pulse.

Pulse is a first-of-its-kind AI intelligence solution that continuously monitors portfolios, market news, regulatory filings, and client CRM data. Built to understand positions, restrictions, tax lots, and workflows, Pulse does more than summarize; it helps users move from insight to action within the Vestmark platform.

Watch the demo here 👇

#WealthManagement #WealthTech #FinancialAdvisors

[Freedom Dumlao @APIguy]

@GregMolnar Well we did used to call them "memory banks"

[Freedom Dumlao @APIguy]

@pcai_ Welcome back, and congratulations!

[Freedom Dumlao @APIguy]

@GregMolnar Lets go! Looking forward to seeing you there!

[Freedom Dumlao @APIguy]

I have been trying to say this.

[Ludonauta @LudensLudonauta]

I was studying MCPs and this comment really blew my mind.

[Freedom Dumlao @APIguy]

@robbyrussell @zeeg Ahhhh and I got sniped by it😔

[Freedom Dumlao @APIguy]

@JasonSwett Carpenters motto:
"Do your best and caulk the rest"

[Freedom Dumlao @APIguy]

@JasonSwett I always call things what they are, but literally:

First integer I define? integer_1

[Freedom Dumlao @APIguy]

@JasonSwett "bug creation is free now"

[Freedom Dumlao @APIguy]

@rushing_andrei I had this thought right after the Fable takedown. Reminded me of the days when exporting any encryption greater than 40bit was a crime

[Freedom Dumlao @APIguy]

Local models are actually pretty good now.

No, they aren't a reliable source of facts and the baked in knowledge is pretty rough. Done use them for what they "know

Use them for what they can "do"!

They are great for using tools, managing workflows, producing structured outputs, condensing and summarizing text, classification and so much more.

If you install a local model and ask it to recite some facts or tell you about some historical concept it will likely fail. Give it some tools, some memory, and a harness though and WOW

[Freedom Dumlao @APIguy]

Anyone who knows me knows I've had an obsession with DSLs for as long as I've been writing code. The idea of a shorthand language for expressing the solution to specific problems is so powerful. And now in the world of AI there is an entirely new reason for DSLs to exist!

[Dave Blundin @DaveBlundin]

The world is going to be desperately short on compute for the rest of time. No matter how many chips we make, demand will outrun supply. Scaling models in parameter count, context length, and thinking tokens continues to yield increasingly better models with consistently slower, more costly responses.

This points to a future with giant frontier models doing some things, but the vast majority of tokens will run through thousands of small, fine-tuned models. Some may even operating with their own domain-specific language. There’s no technical limitation preventing a fine-tuned Gemma 12B model from outperforming Opus 4.8 at a specific task, like linting. But it would run at less than 1% of the cost.

Everyone keeps telling young founders, "AI will do everything, there's no purpose in life." Wrong. Domain-specific models and domain-specific languages will be a goldmine in the next few years.

[Freedom Dumlao @APIguy]

@cheuk_baby @DaveBlundin It's true! But you have options.

[Freedom Dumlao @APIguy]

What you get from the LLM providers is wrapped in guardrails which are limiting what you can do. Doubly true if you access the models via the browser "chat" interface.

[Dave Blundin @DaveBlundin]

The leading LLMs are so guardrailed. So, how do you deploy AI into sensitive areas like healthcare, biotech, and financial services?

Bypass the harness and go straight to the model via Bedrock, says Freedom Dumlao (@APIguy). I’ve made the transition myself already...mostly so I can easily log to S3 and keep my completions from Sam/Dario - but this is an added benefit!

[Freedom Dumlao @APIguy]

@DannyOkec Like getting a glimpse of the future only to have it ripped away.

[Freedom Dumlao @APIguy]

Is this how the Fable ends?

[Freedom Dumlao @APIguy]

Another amazing conversation with @DaveBlundin getting into some of the nitty gritty details of making AI accessible to everyone and a lot more. What a blast as always, and thank you for having me on the show! 🚀 💪

[Dave Blundin @DaveBlundin]

Just posted Round 2 with Freedom Dumlao (@APIguy), CTO & CAIO at @Vestmark, and somehow the conversation got even better than the first one!

A few weeks ago, Freedom told us how all 400 Vestmark employees have access to Claude Code. In this episode, learn about the freshly installed rack of Mac Minis now running AI agents at Vestmark, why running inference through AWS Bedrock instead of the https://t.co/HHLzuu3ME0 API is an unlock for enterprise, and why small fine-tuned models like Gemma 12B are about to eat a huge chunk of the enterprise AI workload.

Thank you Freedom for returning to the Link Ventures podcast studio!

[Freedom Dumlao @APIguy]

Fine tuning small models is the skill most engineers are still missing from their tool belt. If you aren't learning how to do this yet you are way behind.

[Akshay 🚀 @akshay_pachaar]

Google just dropped a new LLM!

You can run it locally on just 8GB RAM.

Let's fine-tune this on our own data (100% locally):