Andrew B. Martin

Software & Agents || Over 50M users served || Product Experimentation & Marketing

[Andrew B. Martin @andrewbrownmart]

@hyponymous Reality is that which, when you look closer at it, rewards you with more reality

[Andrew B. Martin @andrewbrownmart]

"-- Avg. token spend on @tryramp
increased 21x over the last 12 months"

How much of this is engineers shipping product versus a large scale coordination problem?

[The All-In Podcast @theallinpod]

Chamath: Tokenmaxxing on the top frontier models is “a bridge to nowhere” and a “money-burning furnace”

@chamath reacting to Ramp CEO @eglyman on @CNBC saying:

-- Avg. token spend on @tryramp increased 21x over the last 12 months

-- So, the team built a tool for customers to track it

“He's saying something so important there, because if your engineers are going off randomly in an unguided system and then just ripping through a million tokens at $56, what he's talking about is the eventual downstream impact to earnings.

And that eventually, a bunch of these public market CFOs are gonna show up to Wall Street and they will have missed earnings because their OpEx, at some point, if things are 21x-ing every few months, somebody's going to miss a quarter.

Eric would not have released this Ramp product unless CFOs were like, ‘I can't control the spend.’ And then he's like, ‘Well, here, let me build it for you.’

And then, if enough CFOs essentially turn that feature on and start to rate limit how it's spent, because maybe they're not getting the ROI, and the engineer doesn't care about ROI.

The engineer's like, ‘I want to use the latest, greatest model.’

And maybe you don't need it. Maybe Mira's right, and for 95% of the tasks you should be at one level lower, especially when it costs 1/100th of the cost.

Unless you get a control of this and you can directly say how much money you're making, this is a bridge to nowhere. It is a money-burning furnace.”

[The All-In Podcast @theallinpod]

Chamath: Tokenmaxxing on the top frontier models is “a bridge to nowhere” and a “money-burning furnace”

@chamath reacting to Ramp CEO @eglyman on @CNBC saying:

-- Avg. token spend on @tryramp increased 21x over the last 12 months

-- So, the team built a tool for customers to track it

“He's saying something so important there, because if your engineers are going off randomly in an unguided system and then just ripping through a million tokens at $56, what he's talking about is the eventual downstream impact to earnings.

And that eventually, a bunch of these public market CFOs are gonna show up to Wall Street and they will have missed earnings because their OpEx, at some point, if things are 21x-ing every few months, somebody's going to miss a quarter.

Eric would not have released this Ramp product unless CFOs were like, ‘I can't control the spend.’ And then he's like, ‘Well, here, let me build it for you.’

And then, if enough CFOs essentially turn that feature on and start to rate limit how it's spent, because maybe they're not getting the ROI, and the engineer doesn't care about ROI.

The engineer's like, ‘I want to use the latest, greatest model.’

And maybe you don't need it. Maybe Mira's right, and for 95% of the tasks you should be at one level lower, especially when it costs 1/100th of the cost.

Unless you get a control of this and you can directly say how much money you're making, this is a bridge to nowhere. It is a money-burning furnace.”

[Cursor @cursor_ai]

We had a team of agents rebuild SQLite from its 835-page manual.

It created a replica in Rust which passed 100% of a held-out test suite.

Interestingly, cost varied 15x depending on which model mix we used.

[Andrew B. Martin @andrewbrownmart]

demo is here - https://t.co/znOJI3uZb6

[Andrew B. Martin @andrewbrownmart]

Exploring the dynamics of adding more developers to a code base under agentic development.

With high code throughput very easy for time spent reconciling or coordinating changes to dwarf development time.

[Andrew B. Martin @andrewbrownmart]

Maybe big gains to be had from getting naming right since it can reduce reconciliation time!

https://t.co/BJH9sy3pog

[Ben Vinegar @bentlegen]

👋 New blog post:

"How coding agents read files – and how to write for them"

In it I explore how code optimized for agent-discovery may lead to major token savings on search + retrieval, even in code generated from frontier models like GPT-5.6 Sol

Plus other findings ⬇️

[Andrew B. Martin @andrewbrownmart]

@bentlegen Thanks for writing. I wonder how this relates to agent performance gains from keeping logical branching in the code as simple as possible.

[Max Prime @MatrixPrime_]

Imagine having over 100 Million+
And niggas worried about @dom_lucre 30K

BUT @nikitabier needs to eat

[Stephen Lecce @Sflecce]

Canada’s first-ever license to build a grid-scale Small Modular Reactor has been issued — and it’s happening right here in Ontario.

The Darlington SMR will be the first in the G7, powering our province with clean, reliable, and affordable energy.

Read my full statement ⬇️

https://t.co/mHvSNCYmi5

[Andrew B. Martin @andrewbrownmart]

A good example where it pays to not be fully async

[Ryan Singer @rjs]

If you want buy-in on a shaped concept, async is the least productive way. Why? Because you can't see exactly where in the concept the other party breaks with you.

When you walk through it live, you're together: "okay, yes, yes, uh huh, okay..." and then 💥 boom, a "no" comes.

Now we're in the same context looking at that part that doesn't work or where we disagree. We're aware of which parts we do agree on, that aren't contentious. That's a very good setup to do the work of a negotiation, clarification, or trade-off together.

[Ryan Singer @rjs]

If you want buy-in on a shaped concept, async is the least productive way. Why? Because you can't see exactly where in the concept the other party breaks with you.

When you walk through it live, you're together: "okay, yes, yes, uh huh, okay..." and then 💥 boom, a "no" comes.

Now we're in the same context looking at that part that doesn't work or where we disagree. We're aware of which parts we do agree on, that aren't contentious. That's a very good setup to do the work of a negotiation, clarification, or trade-off together.

[tobi lutke @tobi]

@hunkybill @Nelisa_Rojas @ddebow It's extremely easy to spot: Every one of them has zero success exporting which is the true merit test of every company. If you aren't exporting, but somehow own the domestic market, you are getting help. At best help from inertia, but... unlikely.

[Sakana AI @SakanaAILabs]

The AI Scientist Generates its First Peer-Reviewed Scientific Publication

We’re proud to announce that a paper produced by The AI Scientist-v2 passed the peer-review process at a workshop in ICLR, a top AI conference.

Read more about this experiment → https://t.co/50p2t9tgHC

To our knowledge, this is the first fully AI-generated paper that has passed the same peer-review process that human researchers go through. The paper was produced by an improved version of the original AI Scientist, called The AI Scientist-v2. We’ll be sharing the full details of v2 in an upcoming release.

We conducted this experiment with the full cooperation of both the ICLR leadership and the organizers of the ICLR workshop, @ICBINBWorkshop. We (@_yutaroyamada @cong_ml @shengranhu @RobertTLange) proudly collaborated with UBC (@jeffclune) and Oxford (@FLAIR_Ox) on this exciting project.

[Andrew B. Martin @andrewbrownmart]

@buildcanada The big underlying question is do we need to support youth more in getting started building or increase the payoff from building

[Andrew B. Martin @andrewbrownmart]

Love this conversation.

Why not use the additional income from OAS reform to make the first $250K income for Canadians tax free?

Simpler to administer and more building oriented.

[Build Canada @buildcanada]

Our children are our future. Yet, young Canadians face significantly worse economic prospects than previous generations. Student debt has reached an average of $28,000 per graduate. Home prices have increased 375% since 2000 while incomes have grown only 93%.

This generational divide is stark: measures of happiness place Canada 8th in the world for those over 60, but only 58th for those under 30.

As a result, we’re seeing our future literally shrink away: many young Canadians are delaying or entirely forgoing having children, and the birth rate continues to plummet (down to 1.26 children per woman in 2023 – a similar level to Japan).

At the heart of this crisis is a profound misalignment of priorities. Our social spending heavily favours older people at the expense of younger generations. The federal government spends ~$3 on seniors for every $1 spent on children, despite poverty rates being higher for children.

Our old-age programs were designed for a different time. When Old Age Security (OAS) was introduced in 1952, the average life expectancy was 69 years. Today it is 83.

OAS, which is funded entirely from general taxpayer revenue, is projected to increase from $55 billion in 2024 to over $90 billion by 2030. This rapid growth is happening while the very generations expected to fund these payments are facing unprecedented economic challenges.

This is not only a dire economic crisis but a failure of the social contract. A society that consistently prioritizes present consumption over future investment will inevitably consume its own future.

We can change this. We propose a new savings account that provides every Canadian child with $10,000 at birth, followed by annual contributions throughout childhood, generating approximately $50,000-$60,000 by age 18.

Young Canadians could access 25% of the funds at 18, 21, 25, and 28. Access at earlier ages would be permitted for qualified education, addressing essential needs, while leaving the flexibility to use funds however they wish at later ages. Any unused funds can remain invested tax-free for as long as they choose.

We can fully fund this program through modest OAS reforms by raising the eligibility age to 67 and enhancing means testing.

At a time when housing and education costs threaten to lock the next generation out of Canada, this is an earnest step towards renewing the social contract. It has the opportunity to foster independence and prepare our children to be confident, financially literate adults. More importantly, it’s an investment into the most critical part of our future: our children.

Read the full memo at the link in thread:

[Eric Gu @ericyuegu]

I trained a 20M param Transformer on 3 billion frames of Super Smash Bros. Melee replays. No RL, pure behavior cloning.

It wins 95% of the time against level 9 CPU, and only took about $5 to train (5 hrs on two 3090s).

[Andrew B. Martin @andrewbrownmart]

@gbrl_dick Maybe the natural course is once we achieve societal prosperity we just chill until things get bad enough we have to grind again

[Sesame @sesame]

At Sesame, we believe in a future where computers are lifelike. Today we are unveiling an early glimpse of our expressive voice technology, highlighting our focus on lifelike interactions and our vision for all-day wearable voice companions. https://t.co/Edp8V8urgC

[Andrew B. Martin @andrewbrownmart]

ok got it working in 3 prompts with extended turned on

[Ozgur Ozer @ozgrozer]

OMG Claude Sonnet 3.7 is insane. I got this stunning 3D city with one shot. Look at the shadows and the way the day transitions. This is just awesome.