https://www.semianalysis.com/p/google-we-have-no-moat-and-neither?utm_source=substack&utm_medium=email

“Fascinating article - a leaked internal doc from a Google researcher stating that both Open AI and Google will lose the #Artificial Intelligence race to open source.

Who benefits? Amazingly, Meta, because their leaked model is what the open source community has been building off of” h/t @Alex 

——————

Amazing doc. Not sure what I think about it but definitely agree with "LoRA is an incredibly powerful technique we should probably be paying more attention to." ;)

Sam mentioned that GPT4 is result of 100s of manual/proprietary tweaks which seems like small moat tho.  

While I believe communities often beat companies, open vs closed is in a sense like Android vs iOS. You could argue the one who control both hardware and software architecture will control the user experience better and in turn fine tune the AI. And companies have the capabilities to do just that….

Google "We Have No Moat, And Neither Does OpenAI"

Artificial Intelligence

Alex

https://www.semianalysis.com/p/google-we-have-no-moat-and-neither?utm_source=substack&utm_medium=email  “Fascinating article - a leaked internal doc from a Google researcher stating that both Open AI and Google will lose the #Artificial Intelligence race to open source.  Who benefits? Amazingly, Meta, because their leaked model is what the open source community has been building off of” h/t @Alex   ——————  Amazing doc. Not sure what I think about it but definitely agree with "LoRA is an incredibly powerful technique we should probably be paying more attention to." ;)  Sam mentioned that GPT4 is result of 100s of manual/proprietary tweaks which seems like small moat tho.    While I believe communities often beat companies, open vs closed is in a sense like Android vs iOS. You could argue the one who control both hardware and software architecture will control the user experience better and in turn fine tune the AI. And companies have the capabilities to do just that….

Since DeepSeek’s release, Meta has been ramping up efforts to make its open-source AI model, Llama, a serious contender—courting enterprise customers and even the U.S. government. The company is now building a go-to-market strategy that could mirror business models used by other open-source AI firms.
But questions remain about whether Llama can catch up to competitors like DeepSeek, which has focused on efficiency and cost-effectiveness from the start.

https://www.theinformation.com/articles/meta-scrambles-after-chinese-ai-equals-its-own-upending-silicon-valley?offer=exp-ann-25&utm_campaign=Expansion%3A+Meta+Llam&utm_content=7201&utm_medium=email&utm_source=cio&utm_term=4924&rc=gzsl71

 #The Information

The Information

Meta’s AI teams are freaking out about rival AI model DeepSeek’s advances. To figure out how DeepSeek did what it did, Meta’s teams started four war rooms to learn how the model works. 🍿


https://www.theinformation.com/articles/meta-scrambles-after-chinese-ai-equals-its-own-upending-silicon-valley?utm_campaign=article_email&utm_content=article-14402&utm_medium=email&utm_source=sg&rc=gzsl71

Meta’s AI teams are freaking out about rival AI model DeepSeek’s advances. To figure out how DeepSeek did what it did, Meta’s teams started four war rooms to learn how the model works. 🍿   https://www.theinformation.com/articles/meta-scrambles-after-chinese-ai-equals-its-own-upending-silicon-valley?utm_campaign=article_email&utm_content=article-14402&utm_medium=email&utm_source=sg&rc=gzsl71

Artificial intelligence researchers at Meta Platforms have been in panic mode. In recent days, leaders of some of the company’s AI teams openly worried that new conversational AI made by a Chinese hedge fund meant Meta was falling behind in the AI race. Leaders including AI infrastructure ...

Meta Scrambles After Chinese AI Equals Its Own, Upending Silicon Valley

Sam Mendoza

Since DeepSeek’s release, Meta has been ramping up efforts to make its open-source AI model, Llama, a serious contender—courting enterprise customers and even the U.S. government. The company is now building a go-to-market strategy that could mirror business models used by other open-source AI firms. But questions remain about whether Llama can catch up to competitors like DeepSeek, which has focused on efficiency and cost-effectiveness from the start.  https://www.theinformation.com/articles/meta-scrambles-after-chinese-ai-equals-its-own-upending-silicon-valley?offer=exp-ann-25&utm_campaign=Expansion%3A+Meta+Llam&utm_content=7201&utm_medium=email&utm_source=cio&utm_term=4924&rc=gzsl71   #The Information

Jerome Cody

the models are open source, so everyone is free to use the same techniques going forward. And lower costs also mean lower barriers to entry, while more market entrants could increase demand for chips. This is Jevons paradox, where improved efficiency of a resource increases consumption of it.

https://www.scmp.com/business/china-business/article/3296562/deepseek-breakthrough-raises-investor-concerns-over-need-massive-ai-spending?utm_medium=email&utm_source=cm&utm_campaign=enlz-insidechinatech&utm_content=20250207&tpcc=enlz-insidechinatech&UUID=f9aaf50d26626de042873bfe2733d91d&tc=8

the models are open source, so everyone is free to use the same techniques going forward. And lower costs also mean lower barriers to entry, while more market entrants could increase demand for chips. This is Jevons paradox, where improved efficiency of a resource increases consumption of it.  https://www.scmp.com/business/china-business/article/3296562/deepseek-breakthrough-raises-investor-concerns-over-need-massive-ai-spending?utm_medium=email&utm_source=cm&utm_campaign=enlz-insidechinatech&utm_content=20250207&tpcc=enlz-insidechinatech&UUID=f9aaf50d26626de042873bfe2733d91d&tc=8

Large US tech companies may follow suit and replicate some of the training techniques that DeepSeek leveraged to drive the cost of R1 down.

DeepSeek advance triggers Nvidia rout, raising concerns over AI spending

OpenAI CEO Sam Altman seemed to acknowledge the DeepSeek challenge this week and said on Reddit that the company is considering releasing some model weights—or the settings that determine how a model responds to questions—publicly.

“I personally think we have been on the wrong side of history here and need to figure out a different open source strategy,” he wrote.

https://www.theinformation.com/articles/chatgpt-subscribers-nearly-tripled-to-15-5-million-in-2024?utm_campaign=%5BREBRAND%5D+%5BTI-AM%5D+Th&utm_content=1095&utm_medium=email&utm_source=cio&utm_term=124&rc=gzsl71

OpenAI CEO Sam Altman seemed to acknowledge the DeepSeek challenge this week and said on Reddit that the company is considering releasing some model weights—or the settings that determine how a model responds to questions—publicly.  “I personally think we have been on the wrong side of history here and need to figure out a different open source strategy,” he wrote.  https://www.theinformation.com/articles/chatgpt-subscribers-nearly-tripled-to-15-5-million-in-2024?utm_campaign=%5BREBRAND%5D+%5BTI-AM%5D+Th&utm_content=1095&utm_medium=email&utm_source=cio&utm_term=124&rc=gzsl71

Paid subscribers to ChatGPT nearly tripled to 15.5 million last year from 5.8 million a year earlier, OpenAI recently told some shareholders, despite competition from chatbots made by Google, Anthropic and Meta Platforms. Based on what OpenAI charges for the chatbot subscriptions, the increase ...

ChatGPT Subscribers Nearly Tripled to 15.5 Million in 2024

Cyril

The Trump presidency will determine the future for open-source AI.

https://www.theinformation.com/articles/the-battle-over-open-source-ai-will-be-settled-in-trumps-presidency?offer=rtsu-engagement-24&utm_campaign=Expansion%3A+AI+Agenda&utm_content=5978&utm_medium=email&utm_source=cio&utm_term=4075&rc=gzsl71

The Trump presidency will determine the future for open-source AI.  https://www.theinformation.com/articles/the-battle-over-open-source-ai-will-be-settled-in-trumps-presidency?offer=rtsu-engagement-24&utm_campaign=Expansion%3A+AI+Agenda&utm_content=5978&utm_medium=email&utm_source=cio&utm_term=4075&rc=gzsl71

D J

A group of likeminded people in Artificial Intelligence are talking about this.