Don

Data infra

[Bobby Samuels @BobbySamuels]

https://t.co/0muCOoj0BN

[Joshua Kushner @JoshuaKushner]

stay focused...ignore the rat race...try to build things that are cool. the rest will fall into place

[Don @DMFA19]

@jimfarley98 Clearly the coyote v8 can fit in the current chassis...Let's GOOOO...When's the Raptor-R dropping???

[Don @DMFA19]

@Ford Some of us are trying to buy your cars. When is your backup cam update coming out??? There's a LOT of parked cars on lots, waiting to be purchased....

[Don @DMFA19]

@jimfarley98 It would be easier to buy your cars if they weren't under recall. Literally trying to give @ford my money and you can't take it bc you can't push out a software update for a backup cam in under 3 months??? Fix it, @jimfarley98

[Don @DMFA19]

@sundeep 💯

[Tsarathustra @tsarnick]

https://t.co/lp0f0B3bqS says AI will be an amplification tool for creators, giving rise to "the age of the ideator" where ideas are seeded by humans and multiplied by AI systems

[Don @DMFA19]

���️♥️♥️

[Neil Jacobs @NeilJacobs]

Today seems like a good day to rewatch this Michael Saylor video from 2012 about Apple stock. 🍎

He also briefly talks about Dell computers. 👀

#Bitcoin

[Don @DMFA19]

Attention to detail is one thing. Taking in all the details, creating, and executing strategically bassed on those details...that's the stuff the very best players and teams do...repeatedly

https://t.co/bGIRi145e3

[Don @DMFA19]

Ridiculous

[Martin Bowling @martinbowling]

🚀 Introducing YT Groqscriber: Transcribe YouTube videos at warp speed! 🎥➡️📝
Powered by @GroqInc's blazing fast Whisper v3, this @Replit template turns videos into text 140x faster than you can say "transcription"!

[Don @DMFA19]

Sonnet 3.5 blowing up the internet
You're either hyped about the possibilities or lamenting the myriad reasons all of this is "overhyped"..
There are real problems to solve before we get to any sort of panacea...but, damn, the innovation curve is moving ridiculously fast

[Don @DMFA19]

Big things happen every time we meet...

[Don @DMFA19]

@sundeep I grew up on the Detroit side. Gordie Howe international looking nice!!!!

[Don @DMFA19]

Preach!!

[Aaron Levie @levie]

One of the biggest open questions with AI is its impact on software business models over time. What seems to be under-appreciated about AI is how it can enable significant TAM expansion for a large number of categories over time when software can deliver outcomes, and not just enable existing work.

Right now the dominant business model in SaaS is a per seat model, which inevitably means that the total number of seats you can sell is limited to the number of employees in the organization that are relevant for your particular software. Legal software is roughly capped by the size of the legal team, audit software is capped by the size of the audit team, and so on. The implication of this is that the customer generally *already* has to have not only a need for your solution, but also the existing headcount in the organization to become users of your software. Incidentally, this is often why so many SaaS products tend to go after horizontal productivity categories, because this maximizes the number of potential users you have access to in an organization.

AI flips this on its head, especially with the power of AI Agents, and you get a new form of “outcome-as-a-service”. When AI is actually doing the work within the software itself, you're no longer constrained by the number of employees inside the organization to use the actual software. The software is quite literally bringing along the work with it and delivering a particular business outcome. It's clear the full potential of this playing out is not fully understood, as this represents a massive transformation of software as an industry. 

When you are no longer constrained by the size of a team or department to use your software, markets are no longer arbitrarily capped in size. In this new era, software that powers a legal workflow actually brings the equivalent of legal knowledge work along with it, and software for audits brings the equivalent of audit work with it. All of a sudden small businesses, under-resourced teams in large enterprises, and all new geographies begin to open to up as markets. AI will enable otherwise niche categories of software to become much larger, and already large categories of software to become even bigger.

This transformation is similar to what we've seen in other markets where a new innovation has unlocked the size of a market well beyond its original demand. For instance, most investors and economists would have thought the size of Uber or Lyft's market was the size of the existing Taxi market, when in fact the market size was orders of magnitude larger once the shape of the product changed to make the offering easier to consume. We’ve seen this effect time and time again in areas like SaaS, cloud computing, a variety of mobile categories, and more.

We’re only in the earliest of stages of figuring out what this all means for the future of the software business model. Clearly all new variables of monetization will need to emerge when you start to pay software vendors for outcomes as opposed to just the software itself. But inevitably, when you remove the existing dominant constraint of enterprise software, TAM expansion will follow.

[Don @DMFA19]

Interesting recap here...

[Zach Vorhies / Google Whistleblower @Perpetualmaniac]

My god, this paper by that open ai engineer is terrifying.

Everything is about to change.

AI super intelligence by 2027.

[Don @DMFA19]

👀👀👀👀👀
https://t.co/Tnv9y2WI3Y

[Don @DMFA19]

Great example of structure and process gone wrong...to devastating effect...

[Mihail Eric @mihail_eric]

How Alexa dropped the ball on being the top conversational system on the planet
—
A few weeks ago OpenAI released GPT-4o ushering in a new standard for multimodal, conversational experiences with sophisticated reasoning capabilities.

Several days later, my good friends at PolyAI announced their Series C fundraising round after tremendous growth in the usage of their enterprise voice assistant.

Amid this news, a former Alexa colleague messaged me: You’d think voice assistants would have been our forte at Alexa.

For context, I joined Alexa AI as a research scientist in early 2019. By this time, the Alexa consumer device had existed for 5 years and was already in 100M+ homes throughout the world.

In 2019, Alexa was experiencing a period of hypergrowth. Dozens of new teams sprouted every quarter, huge financial resources were invested, and senior leadership made it clear that Alexa was going to be one of Amazon’s big bets moving forward.

My team was born amidst all this with a simple charter: bring the latest and greatest in AI research into the Alexa product and ecosystem. I’ve often described our group (later dubbed the Conversational Modeling team) as Google Brain meets Alexa AI-SWAT team.

Over the course of the 2.5 years I was there, we grew from 2 to ~20 and tackled every part of the conversational systems stack.

We built the first LLMs for the organization (though back then we didn’t call them LLMs), we built knowledge grounded response generators (though we didn’t call it RAG), and we pioneered prototypes for what it would mean to make Alexa a multimodal agent in your home.

We had all the resources, talent, and momentum to become the unequivocal market leader in conversational AI. But most of that tech never saw the light of day and never received any noteworthy press.

Why?

The reality is Alexa AI was riddled with technical and bureaucratic problems.

Bad Technical Process
–
Alexa put a huge emphasis on protecting customer data with guardrails in place to prevent leakage and access. Definitely a crucial practice, but one consequence was that the internal infrastructure for developers was agonizingly painful to work with.

It would take weeks to get access to any internal data for analysis or experiments. Data was poorly annotated. Documentation was either nonexistent or stale.

Experiments had to be run in resource-limited compute environments. Imagine trying to train a transformer model when all you can get a hold of is CPUs. Unacceptable for a company sitting on one of the largest collections of accelerated hardware in the world.

I remember on one occasion our team did an analysis demonstrating that the annotation scheme for some subset of utterance data was completely wrong, leading to incorrect data labels.

That meant for months our internal annotation team had been mislabeling thousands of data points every single day. When we attempted to get the team to change their annotation taxonomy, we discovered it would require a herculean effort to get even the smallest thing modified.

We had to get the team’s PM onboard, then their manager’s buy-in, then submit a preliminary change request, then get that approved (a multi-month-long process end-to-end).

And most importantly, there was no immediate story for the team’s PM to make a promotion case through fixing this issue other than “it’s scientifically the right thing to do and could lead to better models for some other team.” No incentive meant no action taken.

Since that wasn’t our responsibility and the lift from our side wasn’t worth the effort, we closed that chapter and moved on.

For all I know, they could still be mislabeling those utterances to this day.

Fragmented Org Structures
—
Alexa’s org structure was decentralized by design meaning there were multiple small teams working on sometimes identical problems across geographic locales.

This introduced an almost Darwinian flavor to org dynamics where teams scrambled to get their work done to avoid getting reorged and subsumed into a competing team.

The consequence was an organization plagued by antagonistic mid-managers that had little interest in collaborating for the greater good of Alexa and only wanted to preserve their own fiefdoms.

My group by design was intended to span projects, whereby we found teams that aligned with our research/product interests and urged them to collaborate on ambitious efforts. The resistance and lack of action we encountered was soul-crushing.

I remember on one occasion we were coordinating a project to scale out the large transformers model training I had been leading. This was an ambitious effort which, if done correctly, could have been the genesis of an Amazon ChatGPT (well before ChatGPT was released).

Our Alexa team met with an internal cloud team which independently was initiating similar undertakings. While the goal was to find a way to collaborate on this training infrastructure, over the course of several weeks there were many half-baked promises made which never came to fruition.

At the end of it, our team did our own thing and the sister team did their own thing. Duplicated efforts due to no shared common ground. With no data, infrastructure, or lesson sharing, this inevitably hurt the quality of produced models.

As another example, the Alexa skills ecosystem was Alexa’s attempt to apply Amazonian decentralization to the dialogue problem. Have individual teams own individual skills.

But dialogue is not conducive to that degree of separation of concerns. How can you seamlessly hand off conversational context between skills? This means endowing the system with multi-turn memory (a long-standing dream of dialogue research).

The internal design of the skills ecosystem made achieving this infeasible because each skill acted like its own independent bot. It was conversational AI by an opinionated bot committee each with its own agenda.

Product-Science Misalignment
—
Alexa was viciously customer-focused which I believe is admirable and a principle every company should practice. Within Alexa, this meant that every engineering and science effort had to be aligned to some downstream product.

That did introduce tension for our team because we were supposed to be taking experimental bets for the platform’s future. These bets couldn’t be baked into product without hacks or shortcuts in the typical quarter as was the expectation.

So we had to constantly justify our existence to senior leadership and massage our projects with metrics that could be seen as more customer-facing.

For example, in one of our projects to build an open-domain chat system, the success metric (i.e. a single integer value representing overall conversational quality) imposed by senior leadership had no scientific grounding and was borderline impossible to achieve.

This introduced product/science conflict in every weekly meeting to track the project’s progress leading to manager churn every few months and an eventual sunsetting of the effort.

—
As we look forward, in the battle for the future of the conversational AI market, I still believe it’s anyone’s game.

Today Alexa has sold 500M+ devices, which is a mind-boggling user data moat. But that alone is not enough.

Here’s how I would organize a dialogue systems effort from the ground-up:

Invest in robust developer infrastructure especially around access to compute, data quality assurance, and streamlined data collection processes. Data and compute are the lifeblood of modern ML systems so proactively setting up this foundation is imperative.

Make LLMs the fundamental building block of the dialogue flows. In retrospect, the Alexa skills ecosystem was a premature initiative for the abilities of conversational systems at the time. I liken it to when Leap Motion created and released a developer SDK before the underlying hardware device was stable.

But with the power of modern LLMs, I’m optimistic about redesigning a developer conversational toolkit with LLMs as their primitives.

Ensure product timelines don’t dictate science research time frames. Because things are moving so fast in the AI world, it’s hard not to feel the pressure of shipping quickly. But there are still so many unsolved problems that will take time to solve.

Of course you should conduct research aggressively, but don’t have delivery cycles measured in quarters, as this will produce inferior systems to meet deadlines.

—
If you’re thinking about the future of multimodal conversational systems and interfaces, I would love to hear from you. We’ve got work to do!

[Don @DMFA19]

Oh, hello there,
@GroqInc
"Lower is better"
also
"Higher is better"
LOL.
#domination

[Marvin Baumann @MarvinTBaumann]

$MSFT @Azure is the most expensive and the slowest on current SoTA Open Source model inference, LLAMA-3 70B. This does not look good. At least throughput should be faster. Competitor's prices might be heavily subsidised.

[Don @DMFA19]

This is a wild story...

[i’m not stable @unstable_dude]

Hacking time to recover $3m worth of lost Bitcoin.

Sounds crazy, right?

This is how two white hackers cracked an 11 year old password behind this massive fortune.. 🧁

[Don @DMFA19]

Kind of a big deal....