Information & Chat > Investment & Business Discussion

Chat GPT, Deepseek and AliBaba AI Models - And AliBaba Shares

(1/2) > >>

Manny:
DeepSeek's Rapid Ascendancy

Chinese AI startup DeepSeek has made significant waves by releasing its R1 model, a 671-billion-parameter open-source reasoning AI. Remarkably, this model was developed with a budget of under $6 million, a fraction of the investment by competitors. Following its launch, DeepSeek's AI assistant quickly became the top free app on the U.S. Apple App Store, surpassing OpenAI's ChatGPT. This achievement has substantially impacted global tech markets, with major companies like Nvidia, Microsoft, and Alphabet experiencing notable stock declines.

Alibaba's New AI Model

In response to DeepSeek's advancements, Alibaba has unveiled its latest AI model, Qwen 2.5-Max. The company claims that this model outperforms existing competitors, including DeepSeek's V3 and OpenAI's GPT-4o, in various benchmarks. Alibaba's cloud division emphasized that Qwen 2.5-Max achieves competitive performance against top-tier models, marking a significant step in the company's AI development efforts.

It seems Ali Baba is a buy. Analysts currently have a consensus "Strong Buy" rating on Alibaba's stock. The average 12-month price target is $121.33, suggesting a potential upside of approximately 26% from the current price.

I dipped my toe in this morning for a couple of grand. Stock ticker BABA.

andrewfi:
WRT development costs, it seems something might have been lost in translation! It seems that the $6,000,000 being touted about is only part of the total development cost. However, it seems that the cost was pretty low.

The big wobbler comes not from the development cost but from the cost to use and that the models are open source.

It seems from those who have tried it that DeepSeek is not great for content creation (my use case). Although it is already available in my tool set I have yet to take it for a spin.

It appears that one can run at least one model with an SBC such as a Raspberry Pi—that seems amazing. I will deffo look into that as I can see some fun uses if true. If the models scale, then running locally on a PC becomes pretty accessible without dumping thousands into an RTX 4090 or 5090.

Of course, all this fine work was stolen from the USA and that is why the U.S. is now perceived as falling behind. ;)

Contrarian:

--- Quote from: andrewfi on January 29, 2025, 05:44:24 AM ---WRT development costs, it seems something might have been lost in translation! It seems that the $6,000,000 being touted about is only part of the total development cost. However, it seems that the cost was pretty low.

The big wobbler comes not from the development cost but from the cost to use and that the models are open source.

It seems from those who have tried it that DeepSeek is not great for content creation (my use case). Although it is already available in my tool set I have yet to take it for a spin.

It appears that one can run at least one model with an SBC such as a Raspberry Pi—that seems amazing. I will deffo look into that as I can see some fun uses if true. If the models scale, then running locally on a PC becomes pretty accessible without dumping thousands into an RTX 4090 or 5090.

Of course, all this fine work was stolen from the USA and that is why the U.S. is now perceived as falling behind. ;)

--- End quote ---


 :coffeeread:

andrewfi:
One small datapoint that might help understand the issues related to DeepSeek.

There are concerns that DeepSeek may have used a process called distillation to train its AI models. That is where a new AI is trained using the models and data that already exist. SOme people are complaining that DeepSeek has done this and whining about it.

OpenAI actually makes its model open for distillation and provides an API to do so. At least they did until earlier today! To utilize OpenAI's data, DeepSeek's owners simply needed to visit the website, establish API access, and voila—the task is completed. The tools are free, and, from what I saw, there are no restrictions on who might use the tools.

I have no way to know if DeepSeek did use distillation tools to access OpenAI data, but from what can be seen, doing so was quite legitimate.

Manny:
I downloaded DeepSeek last night and tried it out. Its OK, but not as good as GPT.

But the Chinese learn fast.

Navigation

[0] Message Index

[#] Next page