Mr Tunji's position on Deepseek, OpenAI, and Meta's Llama is a bit flawed because Deekseek proved to be better or at least on par with most OpenAI models. The new OpenAI model released for 'free' is behind a $200/month paywall. Deepseek is free and shows the world that these things are overpriced. Sam Altman and his cohorts in America are questioning whether Deepseek was trained over OpenAI when they trained OpenAI on public information. Imo I have been rethinking my investment portfolio in the American equity market. If they overpriced tech as important as AI, what else has been overpriced? If a Chinese person built a language model with a fraction of what we thought, imagine what more can be done in AI and other industries, especially with cheaper resources in Africa.
I continue to see comments like this and i've now come to accept that most use cases for AI is still surface, because further testing with large data sets, e.g, a spreadsheet with 1,000 entries would not have anyone having this conclusion but i'm glad to be proven wrong. Time will tell
Mr Tunji bungled this topic and I think it was he’s excitement that got the best of him. 1. Deepseek hype is that it matches all paid models of the closed source and done using a fraction of the cost 2. The gpus are still Nvidia but they r old models becos America has banned the export of the latest gpu to china to maintain leadership 3. Information on the team who built deepseek are out there- the guy built his hedge fund off the back of the compute for his quantum financial models. He built two data centers some years back with a cool name sef…cnt remember now 4. Tunji is right though on the nvidia loss…deepseek still and will still use nvidia. 5. What deepseek did better was to use better training strategy where open ai and other big players kept on throwing hardware compute at the problem like IT admins throw ram and compute at performance issues, deepseek was forced to think out of the box using a process that ensure that the billions of data is categorized into experts and ensure that reasoning is done using a subset of the brain and not whole brain. The big guys used the traditional ways read and write happens with relational databases to approach this lmao. Deepseek just showed us their grasp on the workings of the human brain
Not necessarily a better training strategy but a less power consuming strategy. Nvidia loss is based off deepseek showing the world that you do not need all those high-end ai nvidia gpus to get this working. What it has not shown the world is if it really did and how it has done it. But the world don't really care much if it is free and open source. Only issue is made in China.
@@jamieHSquare Thank you o! The guys were making me out to be anti-China and the comments too followed. The cool-aid Deepseek served seems so strong that people aren't even questioning the thing deeply. For instance, once your prompt is complex and has a large data set, it throws an error. Compared to Gemini or Vertex.
My 2 cent: OpenAI's claim of Deepseek cutting corners and training it's LLM on their data is laughable seeing that they basically trained their models off people's data online without taking permission. How's that for cutting corners. 2. Nvidia is still in strong buy for me because, Deepseek may have lowered the barrier to entry into the AI industry, making it affordable for everyone to build their own AI tools for cheap. As a result Nvidia will sell more GPUs across all price points as against the quantum of sales being focused on the high-end GPUs.
This is what I’ve been trying to explain. Deepseek just made future access harder for everyone. The loopholes will be sealed and priced into future iterations
Good points @aaronugede Nvidia is still very much in play in the GPU market , the DeepSeek shock will push them to tweak their price and market model to accommodate and prepare for future cycles of change and disruptive headwinds we expect in the AI space this year and going forward.
I am bullish Deepseek, because Deepseek basically did what Elon did with rockets. Some may say he started with a Russian decommissioned ICBM missile, but hey he showed us how cheaper it was to build rockets
this is what happens when half educated people talk on topics they know nothing about Everything the guy wearing red pants said is false, deepseek beats or matched all existing LLM current models. Deepseek is superior in almost every benchmark. Hearing his analysis make me question every other topics he has talked about. And meta llama is also open source which deepseek is based on. Sir stick to topic you know.
He seems to come off as pro Open AI and pro Nvidia and I'm sure he has his reasons for that but I saw a report somewhere that Deepseek is starting to train their model on chips made by Huawei. The Chinese have succeeded in disrupting the AI space and we know they're working on cracking the EUV lithography chips business and by the sounds of it it's a matter of when not if and when they're done they'll crash the prices of high end chips as well. So he should tamper his expectations going forward.
@@auwalabdulazeez2884 What’s the point in being pro Open A? I have no dog in the fight. I’m speaking from direct use, not hear say. This is my experience and that of many who actually are using these modules. On disruption, I agree with you and said as much. Deepseek has changed the game completely but this doesn’t mean the module compares to Mariner for instance. Yes, it’s as good as Gemini but let’s say what it is.
Let’s be clear. The thing that has the world shook is the cost of training the module. And no! It doesn’t beat Gemini on every parameter. It does on some but not all. Now also, if you’ve used it, you’d realized it’s very censored. I appreciate your feedback though.
Bruh, you could have done this "a less hash" or a better way. A better way could be: "It’s important to have accurate information when discussing AI models. Unfortunately, the person in red pants provided incorrect details, DeepSeek has consistently matched or outperformed existing LLMs in benchmarks. Given this, I have some doubts about the accuracy of his other analyses." And because you also think he is incorrect on a particular subject then you have doubts on his opinion on other subject is "Some how". I think we should respect each other even on the internet. I come in peace.
@@peteroluwatobiakintaro1171It’s welcome. I’m the guy in red pants and I stand by my experience. I get the hype around Deepseek and it’s easy to get carried away. But all opinions are welcome. Hash or not.
Nvidia has nothing to be scared of so long as deepseek cannot prove how it used very low-end cheap Nvidia processors to get deepseek to work. Deepseek selling point is - free and open source. openai and meta are two different industry niche. what meta is doing with llama is what Microsoft is doing with openai. what is your website address? We do not have power how can we have an ai strategy? The guy on suit just knows all this thing they are saying about using AI in Nigeria is crap . Are you guys saying it is our Year because of the amount of bad Years we had or there are actual events pointing to this?
Give me the drinks under the moonlight and the talk with flow naturally.🤔 This show has got legs but needs rebranding.. Hire a top brand manager now to determine the best USP for the channel and build the brand. Bring in GenZ segment.. They think differently and their thinking pattern gets to the money faster.. I believe.. Feels more like Patitos Gang at the moment.
Love this podcast. It has given me so much insight into the Nigerian economy and capital market.
Thanks 👌🏾
You guys should have sections so I can go straight to the part I want to watch
Yes. Time stamp
30:26
@@idokosoftwaretech9505 Thank you. This was helpful.
Riiight 😅
They want you to watch the whole thing.
Good feedback
Mr Tunji's position on Deepseek, OpenAI, and Meta's Llama is a bit flawed because Deekseek proved to be better or at least on par with most OpenAI models.
The new OpenAI model released for 'free' is behind a $200/month paywall.
Deepseek is free and shows the world that these things are overpriced. Sam Altman and his cohorts in America are questioning whether Deepseek was trained over OpenAI when they trained OpenAI on public information.
Imo I have been rethinking my investment portfolio in the American equity market. If they overpriced tech as important as AI, what else has been overpriced? If a Chinese person built a language model with a fraction of what we thought, imagine what more can be done in AI and other industries, especially with cheaper resources in Africa.
I continue to see comments like this and i've now come to accept that most use cases for AI is still surface, because further testing with large data sets, e.g, a spreadsheet with 1,000 entries would not have anyone having this conclusion but i'm glad to be proven wrong. Time will tell
Great show!
My new fav show
Exactly 💯... I'm loving this already
Loving this podcast...
Mr Tunji bungled this topic and I think it was he’s excitement that got the best of him.
1. Deepseek hype is that it matches all paid models of the closed source and done using a fraction of the cost
2. The gpus are still Nvidia but they r old models becos America has banned the export of the latest gpu to china to maintain leadership
3. Information on the team who built deepseek are out there- the guy built his hedge fund off the back of the compute for his quantum financial models. He built two data centers some years back with a cool name sef…cnt remember now
4. Tunji is right though on the nvidia loss…deepseek still and will still use nvidia.
5. What deepseek did better was to use better training strategy where open ai and other big players kept on throwing hardware compute at the problem like IT admins throw ram and compute at performance issues, deepseek was forced to think out of the box using a process that ensure that the billions of data is categorized into experts and ensure that reasoning is done using a subset of the brain and not whole brain. The big guys used the traditional ways read and write happens with relational databases to approach this lmao. Deepseek just showed us their grasp on the workings of the human brain
But wait! You agreed with everything I said. Abi I didn’t read your comments well?
Look when it comes to conversations like this everyone wants to show off how much they've been reading...
@@tunjiandrews792
Not necessarily a better training strategy but a less power consuming strategy.
Nvidia loss is based off deepseek showing the world that you do not need all those high-end ai nvidia gpus to get this working.
What it has not shown the world is if it really did and how it has done it.
But the world don't really care much if it is free and open source.
Only issue is made in China.
@@jamieHSquare Thank you o! The guys were making me out to be anti-China and the comments too followed. The cool-aid Deepseek served seems so strong that people aren't even questioning the thing deeply. For instance, once your prompt is complex and has a large data set, it throws an error. Compared to Gemini or Vertex.
Good Summary
My 2 cent: OpenAI's claim of Deepseek cutting corners and training it's LLM on their data is laughable seeing that they basically trained their models off people's data online without taking permission. How's that for cutting corners. 2. Nvidia is still in strong buy for me because, Deepseek may have lowered the barrier to entry into the AI industry, making it affordable for everyone to build their own AI tools for cheap. As a result Nvidia will sell more GPUs across all price points as against the quantum of sales being focused on the high-end GPUs.
This is what I’ve been trying to explain. Deepseek just made future access harder for everyone. The loopholes will be sealed and priced into future iterations
Good points @aaronugede Nvidia is still very much in play in the GPU market , the DeepSeek shock will push them to tweak their price and market model to accommodate and prepare for future cycles of change and disruptive headwinds we expect in the AI space this year and going forward.
Wdym by making future access harder when they showed the world that AI can be accessed for cheaper? @tunjiandrews792
New set! Ehen! Nairametrics!!!
I am bullish Deepseek, because Deepseek basically did what Elon did with rockets. Some may say he started with a Russian decommissioned ICBM missile, but hey he showed us how cheaper it was to build rockets
this is what happens when half educated people talk on topics they know nothing about Everything the guy wearing red pants said is false, deepseek beats or matched all existing LLM current models. Deepseek is superior in almost every benchmark. Hearing his analysis make me question every other topics he has talked about. And meta llama is also open source which deepseek is based on. Sir stick to topic you know.
He seems to come off as pro Open AI and pro Nvidia and I'm sure he has his reasons for that but I saw a report somewhere that Deepseek is starting to train their model on chips made by Huawei. The Chinese have succeeded in disrupting the AI space and we know they're working on cracking the EUV lithography chips business and by the sounds of it it's a matter of when not if and when they're done they'll crash the prices of high end chips as well. So he should tamper his expectations going forward.
@@auwalabdulazeez2884 What’s the point in being pro Open A? I have no dog in the fight. I’m speaking from direct use, not hear say. This is my experience and that of many who actually are using these modules. On disruption, I agree with you and said as much. Deepseek has changed the game completely but this doesn’t mean the module compares to Mariner for instance. Yes, it’s as good as Gemini but let’s say what it is.
Let’s be clear. The thing that has the world shook is the cost of training the module. And no! It doesn’t beat Gemini on every parameter. It does on some but not all. Now also, if you’ve used it, you’d realized it’s very censored. I appreciate your feedback though.
Bruh, you could have done this "a less hash" or a better way.
A better way could be:
"It’s important to have accurate information when discussing AI models. Unfortunately, the person in red pants provided incorrect details, DeepSeek has consistently matched or outperformed existing LLMs in benchmarks. Given this, I have some doubts about the accuracy of his other analyses."
And because you also think he is incorrect on a particular subject then you have doubts on his opinion on other subject is "Some how".
I think we should respect each other even on the internet.
I come in peace.
@@peteroluwatobiakintaro1171It’s welcome. I’m the guy in red pants and I stand by my experience. I get the hype around Deepseek and it’s easy to get carried away. But all opinions are welcome. Hash or not.
How come the Naira is gaining at the time the reserve is going down?
Improved liquidity
❤❤❤
Drinks and Mics!!!!!!! 🎉
What's the best app to use when investing in foreign stocks?
Optimus and bamboo
Try Kuda
Robinhood
In Nigeria or in the US?
@@ugodre in Nigeria
Nvidia has nothing to be scared of so long as deepseek cannot prove how it used very low-end cheap Nvidia processors to get deepseek to work.
Deepseek selling point is - free and open source.
openai and meta are two different industry niche. what meta is doing with llama is what Microsoft is doing with openai.
what is your website address?
We do not have power how can we have an ai strategy?
The guy on suit just knows all this thing they are saying about using AI in Nigeria is crap .
Are you guys saying it is our Year because of the amount of bad Years we had or there are actual events pointing to this?
A trillion dollars not a billion dollars was wiped out
Give me the drinks under the moonlight and the talk with flow naturally.🤔
This show has got legs but needs rebranding.. Hire a top brand manager now to determine the best USP for the channel and build the brand.
Bring in GenZ segment.. They think differently and their thinking pattern gets to the money faster.. I believe..
Feels more like Patitos Gang at the moment.
Follow the money Drink Champs