Nikhil Varghese

[Nikhil Varghese @restingforrest]

@allenainie @RichardSSutton It made my day when he casually joined our table. Thanks for the picture!

[Fidji Simo @fidjissimo]

I disagree with Dario on most things, but this is spot on: “At this point, saying that AI will cure cancer is more a cliché than it is inspiring, and most people think it is deceptive. The thing that will work is actually curing cancer.”

The thing that frustrates me is that some people often draw a straight line from smarter models → cures, without acknowledging how much infrastructure is missing in between. We’ve still got a lot of work to do, people.

The regulatory bottleneck gets a lot of attention. But the bigger bottleneck may be that in many places, we’re lacking the right biological data. AI can’t reason its way to cures if we haven’t generated the data required to understand the disease in the first place.

Ironically, that’s why I’m most bullish on AI making dramatic progress in cancer first: decades of investment have produced extraordinary datasets across genomics, pathology, imaging, clinical outcomes, and more.

For complex chronic diseases, much of that infrastructure simply doesn’t exist yet. The good news is that many companies (including @ChronicleBioAI) are racing to build it.

But model intelligence and biological infrastructure are going to have to scale together. I really do believe AI can cure all diseases but only if we build the infrastructure to translate intelligence into cures at the same pace that models improve (which is a high bar!). Otherwise, we risk having superintelligent AI with an incomplete picture of human biology, and delay that promise by years.

[Nikhil Varghese @restingforrest]

@RL_Conference I registered for the conference yesterday. How can I know if will get a Cirque du Soleil ticket? I couldn't find any contact information anywhere

[Nikhil Varghese @restingforrest]

I don't understand why Claude still does not support structured output. Can someone explain?

[Nikhil Varghese @restingforrest]

@piesauce @MeganRisdal Burma love is great!

[Nikhil Varghese @restingforrest]

@shaneparrish Unfortunately, it's really hard for a media outlet to be unbiased in a free system. The game is to feed what their viewers want to hear. And that just leads to more polarity. It makes sense to make a govt funded media outlet that can stay neutral as govts change

[Nikhil Varghese @restingforrest]

@deliprao I don't have it either in Toronto

[Nikhil Varghese @restingforrest]

@srush_nlp @nikhilmulani Looking forward to the videos! Please share on X once they are available.

[Nikhil Varghese @restingforrest]

"Learn about" is genuinely incredible. Highly suggest you give it a go.
https://t.co/HRnzB0Jb2s

[Paul Couvert @itsPaulAi]

I am genuinely impressed

Google has quietly released “Learn about” a genius new AI tool:

You can enter any topic you want to learn and you'll be able to dive right in VERY easily.

It suggests the related aspects, simplifies, illustrates, etc. More below:

[Nikhil Varghese @restingforrest]

@ArtificialAnlys @cerebras That's tremendous! Would be good to know a cost/token comparison as well to have a more rounded picture

[meg.ai 🇨🇦 @MeganRisdal]

I'm giving a talk about Gemma & its growing tooling ecosystem, @kaggle, open models, and maybe even competitive ML game shows tomorrow at noon ET

And it's virtual so you can attend! Add it to your calendar here: https://t.co/zcGj47desd

[Nikhil Varghese @restingforrest]

@deedydas Congrats on the green card and kudos to you for being so transparent. This is going to benefit so many people!

[Nikhil Varghese @restingforrest]

@llama_index @tb_tomaz @neo4j Great to see. But it seems like Relik doesn't have a permissive license for commercial applications

[Nikhil Varghese @restingforrest]

We should demand an update to the @OpenAI Preparedness Framework Scorecard to include “Anthropomorphization and emotional reliance”

Here's why
How come “Anthropomorphization and emotional reliance” is not the first bullet in the Preparedness Framework Scorecard? It feels like the single largest risk in the deployment of this technology, but yet is buried in the Societal impacts section. The detrimental effects of social media on mental health and society in general are well documented at this point, and we should be wary of repeating our mistakes.

As a pragmatist and believer in the power of AI for good, I appreciate that OpenAI is being forthright in calling this risk out. I understand that the forces of capitalism want this technology to be deployed as soon as possible. But at the very least, we should demand that this risk show up as a first class citizen in the Preparedness Framework Scorecard and not slip through the cracks.

[Nikhil Varghese @restingforrest]

@sundar1791 Good to see Apple maps differentiating in these ways. You should check out immersive view on gmaps. That's also pretty good.

Honestly, when I want to check out a place in 3D, I switch over to Google Earth. That experience is magical.

[Anthropic @AnthropicAI]

What should an AI's character be?

Read our post on how we approached shaping Claude’s character: https://t.co/DUWsl2qFMM

[Nikhil Varghese @restingforrest]

@googleaistudio @OfficialLoganK And a larger surface area of the generated text would also be highly desirable.

[Nikhil Varghese @restingforrest]

@OfficialLoganK I love the experience in @googleaistudio! One minor complaint is the UX while reading the 1.5 pro output in real time. It feels impossible to keep up with the tokens. I end up simply waiting until the whole output is done before I start reading.

[Nikhil Varghese @restingforrest]

@googleaistudio @OfficialLoganK Sure! The lowest hanging fruit is if it let me just scroll up while the model keeps streaming a response.