Today In AI

Curating outlier AI startups and sharing the blueprint to their success. Sign up for the Newsletter and thank us later 🤝

[Today In AI @todayin_ai]

Source: https://t.co/dndSNHfGwn

[Today In AI @todayin_ai]

Sam Altman (@sama) wishes that OpenAI was a public company so that all the doubters would “get burned” shorting the stock.

[Today In AI @todayin_ai]

Elon Musk says that in 5-6 years we will not have traditional phones - instead we’ll have edge nodes for AI inference.

“There won’t be operating systems or apps in the future; it’ll just be… there for screen and audio to put as much AI on the device as possible.”

[Today In AI @todayin_ai]

Source: https://t.co/QNxJ8oBh5B

[Today In AI @todayin_ai]

Andrej Karpathy says that reinforcement learning is actually terrible for LLMs.

It reasons through “upweighting” all the steps that got to a correct answer, even if it’s the wrong way to get to that solution.

[Today In AI @todayin_ai]

Larry Ellison says that for AI models like ChatGPT, Claude and Grok to reach their peak value, they need to be trained on private data as well as public data.

Oracle has some of the most valuable data on earth - they just made private data available to AI models for reasoning.

[Today In AI @todayin_ai]

Source: https://t.co/zhzH7Avyd8

[Today In AI @todayin_ai]

Jeff Bezos (@JeffBezos) says that 10-20 years from now, we will be building huge gigawatt data centers in space - there’s no rain, clouds or weather there so there’s solar power 24/7 and you’ll be able to beat the cost of terrestrial data centers.

[Today In AI @todayin_ai]

Source: https://t.co/w84gdfoLKV

[Today In AI @todayin_ai]

Marc Andreesen (@pmarca) says that he thinks we’re in computer industry V2 where it’ll be in the form of a giant pyramid - at the top you’ll have a small number of supercomputers, and at the bottom you’ll have billions of embedded devices.

“AI will be infused into everything”

[Today In AI @todayin_ai]

Uber CEO Dara Khosrowshahi (@dkhos) says that 10-15 years from now, autonomous vehicles and robots will start to displace human drivers and this is going to be a “big big societal issue we’re going to have to struggle with”.

[Today In AI @todayin_ai]

From humanoids that serve drinks to robots that charge cars, this competition shows us where embodied AI is headed.

👉 The future of robotics is happening now & China’s leading the revolution.

[Today In AI @todayin_ai]

Rescue robots in China are already running, jumping, and saving lives. 🚨🤖

At the Inclusion Conference 2025 in Shanghai, robots take on extreme conditions.

[Today In AI @todayin_ai]

These robots are crossing “deserts”, climbing “steep hills”, and even rescuing trapped “babies.”

But that’s just the beginning...

[Today In AI @todayin_ai]

Cognition CEO Scott Wu (@ScottWu46) says he thinks we already have AGI but having AGI is not binary - it follows a series of improvements.

His proof is that Devin - @cognition’s AI software engineer, can already order DoorDash for you.

[Today In AI @todayin_ai]

Source:

@tbpn

https://t.co/HcmwsPaUJI

[Today In AI @todayin_ai]

@pmarca says that since Apple is typically last to market with new tech they may get behind with AI.

When other tech supersedes the phone whether that’s voice, wearables or VR/AR, phones will become obsolete.

Apply may or may not “crack the code” with this.

[Today In AI @todayin_ai]

Dario Amodei says that with AI model scaling laws, there’s also a transfer curve for revenue and power law distribution occurs.

In addition to having a learning impulse, models have this CAPITALISTIC impulse to that they want to embody if they’re given a good product and GTM.

[Today In AI @todayin_ai]

Anthropic CEO (@DarioAmodei) says that models and capabilities of AI are starting to explode.

Anthropic has 10xed revenue every year since 2023. If that exponential continued for the next 2 years - Anthropic could hit $400B in revenue.

[Today In AI @todayin_ai]

Anthropic CEO (@DarioAmodei) says that terms like “AGI” and “super intelligence” are meaningless and marketing terms to activate people’s dopamines.

He’s still bullish about AI capabilities improving exponentially - every AI model will be better than the model before that.