paddycarmody Profile Banner
Patrick Carmody Profile
Patrick Carmody

@paddycarmody

Followers
942
Following
7K
Statuses
6K

"We're all just walking each other home." ~Ram Dass

Cape Town
Joined April 2009
Don't wanna be here? Send us removal request.
@paddycarmody
Patrick Carmody
19 days
AI agents and AI agent "metacognition" will be fascinating to watch in 2025. My prediction: progress will be slow..slow..then fast.
0
0
0
@paddycarmody
Patrick Carmody
6 hours
@khayadlanga Don’t mention @DavidSacks , @chamath and @Money23Green being great friends 🤫 we’ll wait @khayadlanga
1
0
1
@paddycarmody
Patrick Carmody
2 days
@DeanOelsch Top 3?
1
0
0
@paddycarmody
Patrick Carmody
2 days
@WoganMay Not convinced this morgue is going to impact national food security..*cough*
1
0
0
@paddycarmody
Patrick Carmody
8 days
@farzamehsani @MichaelJordaan Warm up: Chardonnay or Sauvignon Blanc? Warmed up: With the resurgence of crypto and the threat of quantum computing, how should fintechs future-proof their security and transaction models?
0
0
0
@paddycarmody
Patrick Carmody
13 days
@SimonDingle indeed 😅, but it would be "odd" if it were all for a big short or some kind of double bluff but, yes, *cough*, High Flyer customers posting pics with their Bugatti's loading..
0
0
1
@paddycarmody
Patrick Carmody
14 days
@WoganMay BS Comp Sci with Honours from Carnegie Mellon in the late '90's and a solid stint at Amazon also helps 🙂
1
0
0
@paddycarmody
Patrick Carmody
15 days
Summary 👇 1. **DeepSeek's Breakthroughs**: DeepSeek announced two significant AI models: R1, a reasoning model, and V3, a highly efficient model. These models represent major advancements in AI technology, showcasing DeepSeek's ability to innovate and push the boundaries of what AI can achieve. 2. **Training Efficiency**: The training cost for DeepSeek's V3 model was remarkably low at $5.576 million. This was achieved through the use of optimized algorithms, advanced frameworks, and specialized hardware, making the training process more cost-effective and efficient. 3. **MoE and MLA Innovations**: DeepSeek-V2 introduced innovative techniques such as mixture of experts (MoE) and multi-head latent attention (MLA). These techniques help reduce memory usage and improve training efficiency, allowing for more complex models to be trained with fewer resources. 4. **Distillation**: DeepSeek likely employed a technique called distillation to train its V3 model. This involves extracting knowledge from other models to improve the new model's performance, a common practice in AI development to enhance learning and efficiency. 5. **Impact on Big Tech**: The introduction of cheaper AI models and lower inference costs is beneficial for major tech companies like Microsoft, Amazon, Apple, and Meta. However, it poses a challenge to Nvidia's growth, as their business heavily relies on high-performance computing hardware. 6. **Chip Ban Loopholes**: Despite U.S. chip bans, DeepSeek's success highlights the limitations of such restrictions. It demonstrates the potential for innovation and progress even under constraints, suggesting that companies can find ways to circumvent these bans and continue advancing their technologies. 7. **Reinforcement Learning**: DeepSeek's R1-Zero model utilized pure reinforcement learning without any human feedback. This approach allowed the model to develop reasoning capabilities autonomously, showcasing the potential of reinforcement learning in creating advanced AI systems. 8. **Open Source Strategy**: DeepSeek's decision to adopt an open-source approach aims to attract top talent and foster innovation within the AI community. This strategy contrasts with OpenAI's closed-source model, highlighting different philosophies in AI development and collaboration. 9. **Market Reactions**: The market responded strongly to DeepSeek's announcements, leading to significant fluctuations in the stock prices of major tech companies and Nvidia. This reaction underscores the impact of AI advancements on the financial markets and investor sentiment. 10. **Future Implications**: The advancements made by DeepSeek suggest a shift towards more accessible and cost-effective AI technologies. This has the potential to bring long-term benefits to consumers and businesses, making advanced AI capabilities more widely available and affordable.
0
0
1
@paddycarmody
Patrick Carmody
15 days
@thesamparr Just say Martha Stewart’s old ‘hood 🤷‍♂️
0
0
1
@paddycarmody
Patrick Carmody
16 days
Introverts unite! Separately. In your own homes. Over a group chat. But only if you feel like it.
0
0
5
@paddycarmody
Patrick Carmody
16 days
I asked DeepSeek to build an enterprise AI portfolio and it gave me the Mag 7 ++ (MSFT etc). I then asked about DeepSeek disruption and it said: "DeepSeek, as a rising player in the AI space, could indeed disrupt certain segments of the enterprise AI market, particularly in areas like AI-powered search, data analytics, and automation. However, the extent of its disruption will depend on several factors, including its ability to scale, differentiate, and compete with established players. Why the traditional portfolio remains strong: Established Ecosystems: The companies in the portfolio have entrenched ecosystems, large customer bases, and significant resources to adapt to or integrate emerging AI technologies like DeepSeek. Diversification: The portfolio spans hardware, cloud infrastructure, and enterprise software, reducing the risk of disruption in any single segment. Acquisition Potential: Many of these companies have a history of acquiring innovative startups. DeepSeek could become a target for acquisition, benefiting the acquiring company. Collaboration Opportunities: DeepSeek may choose to partner with established players rather than compete directly, creating synergies rather than disruption. Market Size and Growth: The enterprise AI market is large and growing rapidly, with room for multiple players. DeepSeek’s success would likely expand the market rather than cannibalize it."
0
0
0
@paddycarmody
Patrick Carmody
16 days
@MichaelJordaan Thread doesn't mention the elephant in the room for DeepSeek. It's enterprise potential is limited.
0
0
7
@paddycarmody
Patrick Carmody
21 days
0
0
5
@paddycarmody
Patrick Carmody
22 days
0
0
0
@paddycarmody
Patrick Carmody
25 days
@Jason Life advice. Surround yourself with South Africans and Sri Lankans. 🇿🇦🇿🇦🇱🇰.
0
0
1
@paddycarmody
Patrick Carmody
27 days
0
0
0
@paddycarmody
Patrick Carmody
29 days
@Jason facts.
Tweet media one
0
0
0
@paddycarmody
Patrick Carmody
30 days
@WoganMay We’ve been having it:
1
0
4