Sunay Joshi

@sama There are too many degrees of freedom (chat/work, 1.5x speed or not, selecting the thinking effort). Why can't the model set these by itself? Too much burden on user

makes us happy to see people love 5.6 sol so much!

@davisbrownr @HenryKvinge @davisbrownr Some MELD pairs are of form ("express in language of field X", "express in lang of subfield Y of X"); e.g., (modules, vec spaces). Are these as informative as expressing a thm in two *disjoint* fields? E.g., viewing thm abt Brownian motion thru (probability, PDE)?

Some recent work from @HenryKvinge and his collaborators at UW suggests that modest finetuning significantly improves model performance at identifying equivalent mathematical statements expressed in different subfields (e.g., between representation theory and Fourier analysis). This (and related work) suggests some capability overhang, where modest elicitation may lead to models which are better at mathematical or scientific discovery. dlscience.substack.com/p/can-ai-help-…

a famous AI puzzle: since GPT-3, models have consumed a significant fraction of human knowledge. However, only very recent models have made discoveries by connecting ideas across fields. Why did it take so long?

We waited until the blast had passed, walked out of the shelter and then it was extremely solemn. We knew the world would not be the same. A few people laughed, a few people cried. Most people were silent. I remembered the line from the Hindu scripture, the Bhagavad-Gita: Vishnu is trying to persuade the Prince that he should do his duty and, to impress him, he takes on his multi-armed form and says, "Now I am become Death, the destroyer of worlds." I suppose we all thought that, one way or another. -- interview in The Decision to Drop the Bomb (1965)

If you are at ICML, the poster (spotlight) will be at 2pm. Come to say hi and talk with Tuo

Without any private information, I guessed Anthropic might have made some progress in RSI. I think this latest release on the J-space sheds more light. They have a much better sense of the internal representations of the model, which might help crack further improvements

What a special moment at the IMS Annual Meeting in Salzburg. A chamber music concert just for the attendees.

@Cornell @realSuchirJoshi Cool picture

As seen at Cornell: Afternoon light finds its way through the Song of the Vowels sculpture with McGraw Tower looking proudly down.

The Andrew Carnegie Foundation has announced this year’s “Great Immigrants, Great Americans” honorees, including Citi CEO Jane Fraser and Pulitzer Prize-winning authors Hernan Diaz and Cristina Rivera Garza. Read the full story here: buff.ly/legGNg8 #APHighlights

@bremen79 @EdgarDobriban @HamedSHassani ...

Me: LLMs are getting so good they might replace PhD students soon. Him: Actually, there’s an easier job in academia to automate: the PhD supervisor.

He also buys seven pairs of women’s gloves for 17 shillings and gives 1 shilling and 6 pence to charity.

there are surprisingly many open problems when it comes to theory/methods in causal inference check out this talk by Siva Balakrishnan for an excellent & comprehensive summary of the state of the art youtube.com/live/Mnum0Ox1f… stat.cmu.edu/~siva/

For some unknown reason, Thomas Jefferson is preoccupied this morning with recording the temperature. He writes that upon awaking at 6 AM, the temperature is a pleasant 68.0 degrees Fahrenheit, but that it is creeping up toward 70. On his way to Congress, Jefferson purchases a thermometer. It costs three pounds and 15 shillings. No-body knows why he is so concerned with the temperature today, especially, and for what reason he desires an accurate read.

@ml_angelopoulos Amazing work (as usual) @ml_angelopoulos!! We were inspired to apply it to "routing btwn 2 models": arxiv.org/pdf/2605.06479 Controlling rank-instab K was fun; had to bound time a rw w/ drift can dilly-dally in a small region. Your clean write-up made it easy to apply your theory

Today I'm sharing a preprint on conformal risk control for non-monotonic losses, a paper three years in the making. The key idea: validity of conformal can be reframed as a consequence of algorithmic stability. Therefore, any stable algorithm inherits a conformal guarantee. 🧵

@altcap Thank you for this great work Brad!

Dear America,   Starting today, on the 250th birthday of our country, millions of children will now benefit from a new & powerful American birthright - a privately owned investment account from birth. Every child a shareholder in the great American economy! American capitalism is the greatest prosperity-creating engine the world has ever known. It has unlocked unprecedented abundance for our country and untold life-advancing innovation. From clean food & water to cars & planes to vaccines & cures to computers & phones.  Through depressions & world wars, through panics & pandemics, betting on America has been the best investment in human history.  But far too many hardworking families feel like they are on the outside looking in - like the American Dream has become out of reach.  The Invest America Act changes all that. Signed into law by President Trump it establishes the Trump Accounts - launching today - making every child a shareholder from birth.  Every account seeded with $1000 in America's best companies - making every child a direct participant in our stock markets & American capitalism.   But it doesn’t stop there.  Representing the best of our civic traditions, an unprecedented outpouring of additional private  contributions into these accounts is already underway. Billions of dollars have been committed by philanthropists & corporations to tens of millions of children across the country. A new form of massive, direct philanthropy & corporate benefits directly to America’s kids is just getting started. From adopting all the kids of a school or organization to a city or state.  America is the most giving country in the history of the world - and Trump Accounts provide a new open source platform to enable anybody to contribute to anybody.   Trillions of dollars will compound in these accounts over the next 20 years.    But this is about more than just figures on an account statement. By investing in the next generation, our country is providing tangible opportunities to all future generations of Americans: an education, a first home, the ability to start a business, and the foundation for a lifetime of savings. Trump Accounts provide hope to those feeling hopeless - unlocking unlimited human potential. Trump Accounts do this through a massive public - private partnership that empower the individual not the state.  They double down on free market capitalist democracy by empowering the individual rather than growing dependency on government.  Aligning citizens with capitalism rather than pitting them against it.  Perhaps of equal importance, let us not lose sight of the fact that this is a movement that unites our country. Despite the many issues that divide us, this one big idea brings everyone together. From the most conservative to the most progressive leaders, we have found people of goodwill willing to step up and do the right thing to help the next generation.  Trump Accounts represent the very best of our democracy and the American promise issued two hundred and fifty years ago today. I am grateful to all those who have partnered with us to get this far and we dedicate our tireless efforts to make certain that the Invest America Act achieves its maximum potential.  Let us advance into the future with a renewed sense of confidence, optimism and restored faith in the American Dream for EVERY American.    Happy 250th Birthday, America! Brad Gerstner July 4, 2026 @TrumpAccounts @InvestAmerica24 @WhiteHouse @MichaelDell

@Gwynne_Shotwell Amazing!

My husband and I are honored and thrilled to participate in the Invest America program and gift a share of our SpaceX stock to a Trump Account for each of more than two million children across our great nation. Every American child under 18 can benefit from having a Trump Account, but our gift specifically goes to the Accounts of children (ages 11-17) that live in areas with lower average household incomes with a bit more emphasis for those that live near our central Texas home. We have been fortunate in our careers and hope this gift encourages the next generation to continue the journey of enabling humanity to live and fly amongst the stars. A huge thanks to all that worked to bring this incredible program to life.

I am at @icmlconf in Seoul and will be presenting these three works: • Human-AI Collaborative Uncertainty Quantification Tue Jul 7, Poster Session 2 — Hall A, poster 3614 (Also selected for a spotlight talk at the Agentic UQ workshop on Sat Jul 11 in CONF ROOM E1 - E4.) • When to Trust the Cheap Check: Weak and Strong Verification for Reasoning (Spotlight) Wed Jul 8, Poster Session 4 — Hall A, poster 710 • Multiround Human-AI Collaboration with User-Specified Requirements Thu Jul 9, Poster Session 7 — Hall A, poster 3603 Come say hi if you want to chat about any of these works, or in general about human-AI collaboration, uncertainty quantification, verification, or related ideas!!! :)