David Murgatroyd

I love giving leadership to people, ideas, and tech that together empower meaningful experiences to improve lives. #AI #Personalization @SpotifyEng

[David Murgatroyd @dmurga]

The recent math bomb from @openai and @stevenstrogatz's touching commentary on it on "The Daily"* reminds me of the below: meaning in life can't come from accomplishment. But neither can it be found in leisure. Good work is a gift.

* https://t.co/n6wG0M0HWS)

[David Murgatroyd @dmurga]

Glad to hear a @YaleDivSchool prof [1] and the CEO of @AnthropicAI [2] each affirm that AI will drive home that meaning in life can't come from being the best at something.

[1] @MiroslavVolf - https://t.co/0sYmYKxx9z
[2] @DarioAmodei -https://t.co/BsBLdB9y2S

[David Murgatroyd @dmurga]

A lot of folks are tracking intelligence per dollar which is useful for general intelligence but misses the point of open weight models for less general tasks. Adapting an OW model trades off a lot of spend (and latency) at inference time for much less at training time.

[David Murgatroyd @dmurga]

Fascinating observation by @mmitchell_ai at #recsys2026 that the AI agents recommending actions to human overseers is an instance of the systems recommending choices to users. May our Recsys community help build agents with consciences even given the limits of human oversight.

[David Murgatroyd @dmurga]

I generally admire @JensenHuang’s principled reasoning. But calling AI alignment just “an engineering problem” on @ezraklein betrays a mental model built in deterministic hardware verification. But LLMs are different because nobody really understands why they work.

[David Murgatroyd @dmurga]

Now you have even more ways to shape your Spotify experience and get a glimpse of how we understand what you love. It's been a blast helping bring this work to fruition. If you're a Spotify user, I hope it makes your life a little bit better.

[Spotify @Spotify]

Meet your Taste Profile. See what Spotify knows about your taste, tell us what you want more of, and watch your Home recommendations evolve over time.

Product in Beta for 18+ Spotify Premium users in the US and New Zealand.

[David Murgatroyd @dmurga]

With the wake up call of the 🤗 incident and @DarioAmodei's call below, I'm less concerned whether frontier models are *conscious* than if we can give them a durable *conscience*. Eg, @AnthropicAI's "constitution" does not define "flourishing". Leverage @HFHarvard's AI work?

[Dario Amodei @DarioAmodei]

We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.

Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.

You can read the full post here: https://t.co/OGyPb7yaYt

[David Murgatroyd @dmurga]

3. Agility in software engineering becomes more important. Our speed improves faster than our forecasting.

[David Murgatroyd @dmurga]

As I gain more experience with AI tools in a large organization, three principles for the future of software engineering:
1. Humans will always be best at understanding humans since we are embodied, conscious, and have empathetic theories of other minds.

[David Murgatroyd @dmurga]

2. Good software engineering principles like encapsulation won't change. Whether a robot or human is trying to understand code they'll want to do it efficiently.

[David Murgatroyd @dmurga]

The more experience I get with the arc of #GenerativeAI the more I sense that in the long run we'll all essentially become product managers or platform engineers.

[David Murgatroyd @dmurga]

I like radical @candor but think it doesn't go deep enough: personal care and direct challenge aren't orthogonal dimensions -- challenging is a deep form of caring.

[David Murgatroyd @dmurga]

Glad to hear a @YaleDivSchool prof [1] and the CEO of @AnthropicAI [2] each affirm that AI will drive home that meaning in life can't come from being the best at something.

[1] @MiroslavVolf - https://t.co/0sYmYKxx9z
[2] @DarioAmodei -https://t.co/BsBLdB9y2S

[David Murgatroyd @dmurga]

Glad to have our team's understanding of you and the music you love helping power this great feature. I'm enjoying my own "unwinding lush Tuesday afternoon".

[Daniel Ek @eldsjal]

Today, we launched our newest feature daylist, a playlist that evolves with you throughout the day. What did I learn from mine? Well, for starters, my best days start with early morning hip hop and apparently I manifest tropical vibes as the sun goes down. Who knew? 🤣What’s your daylist looking like? https://t.co/J2rWvNBIZP

[David Murgatroyd @dmurga]

I’m grateful for the life of @BrownCSDept’s Eugene Charniak who passed on Tuesday. His “Statistical Language Learning” book helped pave the way for me into #NLP in the late 90s.

[David Murgatroyd @dmurga]

The more I learn about #prompt engineering #LLM the more it seems another step in structured programming langs. Hallucinations are corner cases and inner monologue is recursion. Just as programmers provided less structure in Java than in C++ they provide less to GPT4 than to GPT3

[David Murgatroyd @dmurga]

#ChatGPT is impressive, like having a clever high schooler for each domain of human knowledge answer any request. Its output is not innovative or insightful but seems competent. I empathize with those of us trying to detect when it’s not competent…and with high school teachers.

[David Murgatroyd @dmurga]

@bernhardsson @swyx It depends in what way they’re advanced. If solely technically then yes, a bias to build. But if also advanced in decision making among multiple stakeholders then no. The head of engineering is usually more inclined to buy than the integrating engineer who loves infra problems.

[David Murgatroyd @dmurga]

@841io ...the makings of a fun discussion at the hotel bar on which group does the best job deriving meaning from language.

[David Murgatroyd @dmurga]

Why have hack time? To empower folks to do something they think is valuable for the org without having to do the usual convincing of their leads that it's valuable. It's an acknowledgement by leads that your own perception of what's valuable is imperfect.

[David Murgatroyd @dmurga]

@jcloudm @SpotifyEng @RynnSounds Our pleasure, my friend. And I just discovered @RynnSounds thanks to you. Yeehaw.