Revathi Anilkumar

👩🏽‍💻Data @ @meet_dandy | Prev Apple, Amazon Healthcare | @nyuniversity

[Revathi Anilkumar @RevathiAk25]

https://t.co/394tgxy0Tq

[Visa is doing marketing consults (see pinned!) @visakanv]

I used to get so worked up about this class of phrase (“no one ever told me…” “no one ever said…”) until I started mentally autocorrecting it to “I never listened or noticed when…” and since then I have known peace

[Revathi Anilkumar @RevathiAk25]

My favorite part of a live sporting event is the moment people realize they are on a Jumbotron!

[Revathi Anilkumar @RevathiAk25]

I’ve been lucky to work with Evan on becoming aware of these 5 skills and will admit that I am no where close to mastering them.

But this is also what excites me so much about working with Claude as a coach in parallel to a coach like Evan - helps me reinforce concepts with a lot more rigor

[Evan LaPointe @evanlapointe]

The things you should learn from a coach:
1. Self awareness. Always first.
2. Self control. Eliminate impulsivity.
3. Greater observation. You see less than you think.
4. Understand personalities. They aren’t like you. And can’t be.
5. Learn truth. When to hold your ground.

4.8 is better than 99.99% of coaches at 1-4. But it is firewalled from 5. The tuning is very clear: always create balance and push 1-4. At the pinnacle of learning and leading, 5 must be learned.

But save the money and confusion and use 4.8 for a while so you’re ready for 5. Then, find the 0.01% coach who is a truth knower.

[Evan LaPointe @evanlapointe]

Your deepest skills are thinking well, creating great relationships, and influencing/leading.

Every problem you’ll ever have in your career will stem from a gap in these skills.

Every explanation for why you succeeded where others couldn’t will stem from one of these skills.

[Revathi Anilkumar @RevathiAk25]

This reminds me of @ScottAdamsSays micro lesson on affirmations

https://t.co/sVt4l6KoRi

[Electra @Electrarythm]

You will never outperform your self-image, this is one of the most important things ever said about human behavior and almost nobody understands what it really means, your self-image is the picture you carry inside your head of who you are, what you're capable of, what you deserve, and what's possible for you, and your entire life is just your nervous system executing the orders of that picture, you don't behave according to what you want, you don't behave according to what you say, you don't behave according to your goals, you behave according to who you secretly believe you are, and the gap between where you are and where you want to be is almost always the exact gap between your real self-image and the one you keep trying to talk yourself into.

The plastic surgeon Maxwell Maltz figured this out in the 1950s when he noticed that some patients, even after he fixed their face perfectly, still walked out of his office feeling ugly, and others with minor cosmetic changes walked out feeling brand new, the surgery didn't matter, what mattered was whether the internal picture had changed, and he wrote a book called Psycho-Cybernetics in 1960 that became the foundation of basically every self-development book that came after it, his point was simple, the brain operates like a guided missile that locks onto whatever self-image you've installed, and it will steer you, sabotage you, and bring you home to that image no matter how hard your conscious mind fights, you can win the lottery and end up broke again in two years if your self-image is "poor person," you can lose 50 pounds and gain it back if your self-image is "fat person," you can land your dream job and quietly destroy it if your self-image is "not good enough," because the brain experiences any mismatch between reality and self-image as a problem to be corrected, and it always corrects toward the image.

This is why goal-setting, willpower, motivation, and discipline almost always fail in the long run, they're all happening at the level of behavior while the self-image underneath stays exactly the same, you can't out-discipline a self-image, you can't motivate yourself past it for more than a few weeks before it pulls you back, the only real way to change your life is to change the picture first, and the picture changes through repeated vivid imagination, especially in the relaxed state right before sleep and right after waking, when the critical part of your mind goes quiet and the subconscious actually listens, you spend ten or fifteen minutes a day living inside the version of you you want to become, with full sensory detail, with the feeling of it already being true, you do that consistently for a few months and the internal picture genuinely shifts, and once the picture shifts the behavior follows by itself, no daily battle required, because now your subconscious is steering you toward a different home.

[Revathi Anilkumar @RevathiAk25]

This is fascinating to read. A couple of observations
- the lock in was already there with Claude code having the first mover advantage, and it’s very evident when working at smaller companies with a conservative AI budget. Currently switching costs are high.
- Claude Tag works more like a coworker than Cowork or Claude Code, further justifying that companies allocate their labor spend versus just their IT spend

[Arvind Narayanan @random_walker]

The new Claude Tag feature seems extremely useful, but at the same time, a dangerous bargain for enterprises because of the pricing model and the risk of lock-in. The four big changes together mean that you interact with Claude as a coworker instead of a tool (the same Claude instance for everyone instead of each worker; soaks up tacit knowledge without your telling it; acts on its own; and does so asynchronously). All clearly very useful, but completely flips the interaction paradigm. https://t.co/iWpePXGiL8

Let’s talk about lock-in. As far as I can tell, Claude maintains its own memories in this new way of working; the human team members can’t see and edit them. (System administrators presumably can, but they have other things to do!) Tacit knowledge thus goes from a weakness of AI agents to a major strength — it seems inevitable that as teams and orgs start to use Claude this way, it will become the main queryable repository of all their tacit knowledge, creating dependence and stickiness. Effectively, Claude is a coworker that you can’t fire without *every* team losing workflows and know-how.

By the way, it also seems to introduce a new and pervasive security risk, since Claude can be integrated into private channels as well, and can be given access to repositories and tools even if the users in that channel don’t have access to them. Anthropic has introduced an interesting but complicated access control model to handle all this: https://t.co/l4oB5SVk9r But I’m not sure I trust people to understand and implement it correctly, nor the LLMs to be sufficiently robust against threats like prompt injection.

What about pricing? Claude is not like regular coworkers, because it bills for every token it produces. And it can do an unbounded amount of work, asynchronously and without being asked. In the current model, when AI is a tool, enterprises set per-user budgets, which creates accountability and keeps cost somewhat manageable. When everyone shares a Claude, it will be much harder to track and control spending. Of course you can set a token budget, but turning off Claude for the month for everybody when the budget is hit risks bringing work to a screeching halt.

When AI companies talk about the next stage of AI being a “drop-in replacement” for human workers, it should be understood not as a technical innovation but a business model innovation, enabling more value capture and rent extraction. AI companies are no longer competing for a share of enterprises’ IT budgets but rather a share of their entire labor spend, which is orders of magnitude bigger. Claude Tag is a big milestone in this evolution. This shift is very good for AI companies, but it is unclear if it is good for their customers.

[Revathi Anilkumar @RevathiAk25]

And in the mean time, I plan to keep rewatching The Bear to pick up on these skills. This is your reminder that the final season is out this week🙌

[Revathi Anilkumar @RevathiAk25]

Taking a moment to appreciate the skill and craftsmanship that goes into every meal at a Michelin-starred restaurant

The staff maintains constant communication throughout your meal. If you get up mid meal, they signal to the chef that you’ve stepped away, then signal again when you’re back in your seat. On that cue, the chef begins plating ensuring each dish arrives at your table at the right temperature. And if the staff assigned to your table isn’t available in a given moment, others step in, ensuring the customer always gets a consistent experience held to a high bar of quality.

I aspire to build that level of communication, accountability and foresight skills someday.

[Revathi Anilkumar @RevathiAk25]

@evanlapointe Wishing him a very speedy recovery!

[Akash Ramachandran @akashramzz]

framed wall-hanging gifts for your loved ones

2025: iris scans / night sky on special date
2026: memory graph

[Revathi Anilkumar @RevathiAk25]

3/ Some thoughts to noodle on - i) are there diminishing returns to scaling, ii) by "building context" and introducing human knowledge as context are we limiting these AI agents to ONLY discover what we already have and not discover like we can

[Revathi Anilkumar @RevathiAk25]

3/ Some thoughts to noodle on - i) are there diminishing returns to scaling, ii) by "building context" and introducing human knowledge as context are we limiting these AI agents to ONLY discover what we already have and not discover like we can

[Revathi Anilkumar @RevathiAk25]

Reading Rich Sutton’s Bitter Lesson, I find myself coming back to the constraints I listed out for models to understand taste.

Do the laws of scaling also apply to taste? Is the lesson here that we should stop trying to teach a model taste by encoding what we consider to be tasteful, but rather let the model find its own taste?

[Revathi Anilkumar @RevathiAk25]

4/ But then there are constrains:
- There aren’t that many examples. Tasteful work is rare and you are training on a small sample of data
- Models will do a good job of capturing the aesthetics, but can they capture the reasoning that lead to a specific aesthetic? Maybe possible in the future
- And finally, taste is subjective

[Revathi Anilkumar @RevathiAk25]

2/ Learning is improving from data and experience. Compute takes place ahead of time

[Revathi Anilkumar @RevathiAk25]

6/ My hunch is to solve for taste, we would need to solve for the second constraint. Curious to see how this plays out

[Revathi Anilkumar @RevathiAk25]

Some working notes on taste:

1/ To start with, what is good taste?
For this I look to the one person I know that started talking about Taste before it was mainstream - @shreyas. As he puts it (paraphrasing): there are two parts to taste. First, you can discern what’s good and explain why. The second is timing, where you can spot quality early

[Revathi Anilkumar @RevathiAk25]

5/ Subjectivity may not be a deal breaker though if these models are not targeting “universal taste”