"It may not matter much to OpenAI given its tremendous market momentum and what looks like a technology lead will endure with another model release. It will matter a lot to Microsoft. It is now carving out its own path in generative AI with OpenAI as a pillar, but not the only pillar."
Well put. I wonder if the internal devision, led by Suleyman, will be focusing on improving the core products more than leading the research inside Microsoft. Inflection has shown it can do it, with their models. But with OpenAI's technology becoming instantly available to Microsoft, due to the nature of their partnership, I wonder if Microsoft feels the need to match on that front and really push the frontier in-house in parallel. I also don't see Microsoft's Copilot products ran on Inflection models anytime soon.
I see this under the old investment term of "option value." Option value can be a modest bet with the potential large upside, or a bet designed to limit your downside risk if your primary bets don't materialize as planned. The thinking goes: 1. If OpenAI implodes, I have another in-house bet. 2. I am already working on some vendor risk strategies with small language models (SLM) developed in house (e.g., Phi). 3. However, despite all the hype around SLM, it is the frontier LLMs that are really generating the most value. 4. The key alternative LLM to OpenAI in the market is either from Google or owned by my cloud hyperscaler rivals, Google and Amazon. 5. Mistral might be interesting, but it is located in the EU which almost certainly would cause regulators to take a close look at any tie-up, which is a headache I don't need. 6. Llama could be iinteresting, but there is no deal to be had there. 7. Same consideration for Grok. 8. If I look around at the tier 2 players, nothing particularly distinguishes them. 9. There is a group among the tier 2 where we already have backchannel relationships. 10. Those relationships help me do something creative that enables me to avoid regulatory inquiry, get in-house talent that potentially enables us to chart our own course, and provides a frontier model alternative for use today if required. Maybe this should be it's own post. 😀
"It may not matter much to OpenAI given its tremendous market momentum and what looks like a technology lead will endure with another model release. It will matter a lot to Microsoft. It is now carving out its own path in generative AI with OpenAI as a pillar, but not the only pillar."
Well put. I wonder if the internal devision, led by Suleyman, will be focusing on improving the core products more than leading the research inside Microsoft. Inflection has shown it can do it, with their models. But with OpenAI's technology becoming instantly available to Microsoft, due to the nature of their partnership, I wonder if Microsoft feels the need to match on that front and really push the frontier in-house in parallel. I also don't see Microsoft's Copilot products ran on Inflection models anytime soon.
I see this under the old investment term of "option value." Option value can be a modest bet with the potential large upside, or a bet designed to limit your downside risk if your primary bets don't materialize as planned. The thinking goes: 1. If OpenAI implodes, I have another in-house bet. 2. I am already working on some vendor risk strategies with small language models (SLM) developed in house (e.g., Phi). 3. However, despite all the hype around SLM, it is the frontier LLMs that are really generating the most value. 4. The key alternative LLM to OpenAI in the market is either from Google or owned by my cloud hyperscaler rivals, Google and Amazon. 5. Mistral might be interesting, but it is located in the EU which almost certainly would cause regulators to take a close look at any tie-up, which is a headache I don't need. 6. Llama could be iinteresting, but there is no deal to be had there. 7. Same consideration for Grok. 8. If I look around at the tier 2 players, nothing particularly distinguishes them. 9. There is a group among the tier 2 where we already have backchannel relationships. 10. Those relationships help me do something creative that enables me to avoid regulatory inquiry, get in-house talent that potentially enables us to chart our own course, and provides a frontier model alternative for use today if required. Maybe this should be it's own post. 😀