Leigh Anne Miller

AI/ML. Palo Alto-based. Pr: She/Her

[Nghi Bui @QuocNghi91]

We release OpenDev, a highly configurable coding agent that can scale to hundreds of parallel agents for multiple tasks or a single large coding task. Each session and workflow (Thinking, Compact, Self-Critique, etc.) is highly configurable with models from different providers, making this a highly configurable compound AI system where advanced users can test the quality of different LLMs at the finest granularity. It can also serve as a playground to measure the actual behavior of LLMs on specific workflows.

I have learned a lot by building this. Even though we know "conceptually" how high-quality products such as Claude Code, OpenClaw, and Codex are built, once you get hands-on, there are so many small things to optimize to reach that level of quality. Without actually building one yourself, there is no way to truly learn how these systems work and how to improve them.

More updates, docs, and a website are coming soon.

A long term vision could be an "Operating System for Coding Agents" as a runtime platform that manages the lifecycle, resources, communication, and safety of one or many autonomous coding agents, the same way a traditional OS manages processes, memory, I/O, and security for programs.

More updates, docs and website will coming soon.

Thanks for sharing @omarsar0 !

[Jacob @JacobJaber]

High standards mean nothing without the ability to distinguish between a person who is struggling because the work is genuinely hard, one who hasn’t been given the tools to succeed, and one who simply isn’t willing to meet the bar.

Leaders who can’t make that distinction will either hold the line indiscriminately and lose good people, or capitulate to pressure and watch the standard quietly erode.

The skill isn’t having high standards, it’s maintaining diagnostic clarity under social pressure, developing the people worth developing, and making clean decisions about those who aren’t.

A culture doesn’t degrade because a leader lacked conviction. It degrades because they lacked precision.

[Leigh Anne Miller @LeighAnne3000]

always good to see

[Leigh Anne Miller @LeighAnne3000]

In the late 1950s: Karen Spärck Jones was told her CS Ph.D was uninspired and lacking original thought. SHE INVENTED THE CONCEPT OF INVERSE DOC FREQUENCY (IDF) which is the tech under most modern search engines.#KeepPushing #WomenInTech https://t.co/QZKKHABX7q

[GitHub @github]

Git really is for everyone. Creator Linus Torvalds explains why. 👀

Check out our full interview for Git's 20th anniversary. 👇
https://t.co/qm3ybASwva

[Leigh Anne Miller @LeighAnne3000]

I just learned about the original Bourne shell (Stephen Bourne) and realized most modern shells still maintain backward compatibility with it. #slay

Is it just me... or does learning stuff connecting us to the past programmers make you smile, too? (@Werner)

[Leigh Anne Miller @LeighAnne3000]

@saranormous @pranavreddy @swyx live paper-reading is gonna be my new favorite sorting mechanism for finding other nerds

[Clara Shih @clarashih]

I’m thrilled to share today that I’ve joined @Meta to lead a new Business AI group.  Our vision for this new product group is to make cutting-edge AI accessible to every business, empowering all to find success and own their future in the AI era.

200M businesses each month turn to @facebook @instagram @WhatsApp to connect with billions of consumers around the world. Meta’s Llama models have over 600M downloads to date, and Meta AI now has more than 500M monthly actives— not to mention the incredible ways we'll bring these AI advancements into the physical world through AR glasses and VR headsets. Meta's global reach and leadership in AI represent a generational opportunity for businesses, and I couldn’t be more excited and grateful to help take this from zero to one to scale.

Social media has been the cornerstone of my career. It was the foundation of my startup @HearsaySystems (now part of NASDAQ:@Yext), which enables sales reps to brand themselves as expert advisors on social networks, and my passion for helping business leaders in my 2009 book, The Facebook Era (Prentice Hall).  Today truly feels like a full-circle moment. Now the fun begins!

[litquidity @litcapital]

This October, I’m helping bring Tech Week to CA. To kick things off, we’re giving away 100 limited-edition Lit x Tech Week sweaters and a few invites for exclusive events

To enter:
- like this post
- retweet
- follow both @litcapital and @Techweek_

See you there 🤝

[Not Jerome Powell @alifarhat79]

Linux users right now

[Mac The VC @MacConwell]

Founders, confidence is a must but so is the ability to say you don't know something or having thoughts about what happens if your initial plans don't work out. We love confidence but as a pre-seed company we also know everything won't go perfect, so it's ok to acknowledge that

[Leigh Anne Miller @LeighAnne3000]

In the late 1950s: Karen Spärck Jones was told her CS Ph.D was uninspired and lacking original thought. SHE INVENTED THE CONCEPT OF INVERSE DOC FREQUENCY (IDF) which is the tech under most modern search engines.#KeepPushing #WomenInTech https://t.co/QZKKHABX7q

[Leigh Anne Miller @LeighAnne3000]

In the late 1950s: Karen Spärck Jones was told her CS Ph.D was uninspired and lacking original thought. SHE INVENTED THE CONCEPT OF INVERSE DOC FREQUENCY (IDF) which is the tech under most modern search engines.#KeepPushing #WomenInTech https://t.co/QZKKHABX7q

[Leigh Anne Miller @LeighAnne3000]

@mljmljmlj Best wishes for your healing and we are all rooting for you.

[Tom Holland @holland_tom]

Arrived in Copenhagen, I prepare to join forces with an emotionally repressed but ferociously driven detective inspector, as together we investigate a Nazi serial killer whose crimes will lead us to a confrontation with corruption at the very summit of Danish politics

[Sam Altman @sama]

i'm very happy to welcome our new board members: fidji simo, sue desmond-hellmann, and nicole seligman, and to continue to work with bret, larry, and adam.

i'm thankful to everyone on our team for being resilient (a great openai skill!) and staying focused during a challenging time.

in particular, i want to thank mira for our strong partnership and her leadership during the drama, since, and in all the quiet moments where it really counts. and greg, who plays a special leadership role without which openai would simply not exist. being in the trenches always sucks, but its much better being there with the two of them.

i learned a lot from this experience. one think i'll say now: when i believed a former board member was harming openai through some of their actions, i should have handled that situation with more grace and care. i apologize for this, and i wish i had done it differently. i assume a genuine belief in the crucial importance of getting agi right from everyone involved.

we have important work in front of us, and we can't wait to show you what's next.

[Shriya Nevatia @shriyanevatia]

Who's raising their Pre-Seed or Seed right now?

We would LOVE to hear from you at @TheCouncilCap Angels!!

Pitch us on our website & mention in the comments that you saw this tweet.

Our crew of >150 female operator-angels particularly wants to hear from *female* founders.

[Leigh Anne Miller @LeighAnne3000]

.@karaswisher Burn Book (audiobook) with you reading - could NOT stop listening! Am still in my driveway in the car! Mesmerized! You crushed it!! #burnbook #kararocks

[Leigh Anne Miller @LeighAnne3000]

"When a great team meets a lousy market, market wins.
When a lousy team meets a great market, market wins.
When a great team meets a great market, something special happens." Marc Andreessen, 2007
#startups #GenAI @internetarchive https://t.co/0L8TzTGsXt

[Andrew Ng @AndrewYNg]

It is only rarely that, after reading a research paper, I feel like giving the authors a standing ovation. But I felt that way after finishing Direct Preference Optimization (DPO) by @rm_rafailov @archit_sharma97 @ericmitchellai @StefanoErmon @chrmanning and @chelseabfinn. This beautiful paper proposes a much simpler alternative to RLHF (reinforcement learning from human feedback) for aligning language models to human preferences.

RLHF has been a key technique for training LLMs. In brief, RLHF (i) Gets humans to specify their preferences by ranking LLM outputs, (ii) Trains a reward model (used to score LLM outputs) -- typically represented using a transformer network -- to be consistent with the human rankings, (iii) Uses reinforcement learning to tune an LLM, also represented as a transformer, to maximize rewards. This requires two transformer networks, and RLHF is also finicky to the choice of hyperparameters.

DPO simplifies the whole thing. Via clever mathematical insight, the authors show that given an LLM, there is a specific reward function for which that LLM is optimal. DPO then trains the LLM directly to make the reward function (that’s now implicitly defined by the LLM) consistent with the human rankings. So you no longer need to deal with a separately represented reward function, and you can train the LLM directly to optimize the same objective as RLHF.

Although it’s still too early to be sure, I am cautiously optimistic that DPO will have a huge impact on LLMs and beyond in the next few years.

You can read the paper here: https://t.co/m14qRYszVa I also write more about this in The Batch (linked to below).
https://t.co/8h2ag2plIa