Avatar
arun
66749e7e3301133e813e6d19c8213d5d64a0eabca3ebe3770a232ded96151a2c

Microsoft's Majorana 1 chip advances quantum computing with up to one million qubits, enhancing speed and stability for complex problem-solving.

#microsoft #Quantum #QuantumComputing

https://open.substack.com/pub/deepgains/p/microsofts-quantum-leap-majorana?r=8stv&utm_campaign=post&utm_medium=web

Inference compute is becoming more critical than ever—and one company is making serious waves. Groq is emerging as the one to watch, pushing the boundaries of AI.

#AI #LLM #NVIDIA #GROQ

https://deepgains.substack.com/p/custom-ai-silicon-emerging-challengers

US pushes back against AI regulation, emphasizes Growth over Restrictions - JD Vance at the #AI Action Summit

https://open.substack.com/pub/deepgains/p/us-pushes-back-against-ai-regulation?r=8stv&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

We are at a transition stage where machine still generates code in a human readable form (python, C, shell script etc.) This will soon be unnecessary. #AI

Tools like GitHub copilot, hallmark of this era, are going out once you realize - why even bother with the middleman?

There is a lot of online news articles that talk about Elon Musk's access to US #Treasury. All of them talk about it from #DOGE angle and miss the #XMoney angle totally. Xmoney is is a digital payment platform from his team that was just recently announced.

#ElonMusk

The cost of hosting the flagship non distilled version of #DeepSeekR1 has dropped from 6k to 2k in a week. Without GPU and gives up to 4 tokens per sec.

https://digitalspaceport.com/how-to-run-deepseek-r1-671b-fully-locally-on-2000-epyc-rig

This may be DeepSeek’s biggest win. It has made Sam Altman consider something than even Elon Musk could not make him do. https://cdn.nostrcheck.me/66749e7e3301133e813e6d19c8213d5d64a0eabca3ebe3770a232ded96151a2c/26e566ded540b9f8db9d92feae4f4683b67d1b49bdafd9427e6fce8ea9753382.webp

I expect a muted market reaction to O3 mini release.

Deepseek has already taken all the attention away this week.

O3 is closed source

Charts show a small improvement over O1.

The areas of improvement are super niche. What incremental usecases in real world does this impact?

True power of open source models is the freedom to run it without restrictions. Here is how you can run DeepSeek locally for free and without restrictions.

https://open.substack.com/pub/deepgains/p/running-deepseek-locally-for-free?r=8stv&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

Better and more efficient reasoning models and large language models models are not going to invalidate the need for more investment in compute / high bandwidth memory / data centers and energy. It is only going to demand more of and accelerate the above.

Getting the feeling that Deepseek R1 maybe a psyop.

The output of the model is excellent and the model and weights are open source. Most of the claims will be tested soon. However, the question is primarily on their training compute claims. Scale AI CEO says DeepSeek may be using 50k NVIDIA H100s.

Fediverse is arguably the worst name for a protocol enabling social networking services to communicate using a common protocol. Its origins are from a combination of the words "federation" and "universe"

https://en.wikipedia.org/wiki/Fediverse

Steve Jobs famously said that the computer is like a bicycle for the mind.

Analogously, I think agents are LLMs on a bicycle.

As TikTok goes dark in the US, what are the immediate next steps for marketers ?

There is a possible 90 day extension on the horizon

⬇️ Download content and data.

📣 Let your audience know

🔎 Review agency and influencer contracts

💰 Review, repurpose budget.

♻️ Repurpose the content