I tried out DeepSeek for some research I was doing. It still produces errors and hallucinations like all LLMs. Personally, I didn't find it performed any better than Mistral, with the exception that it accessed more recent data. The "thinking" dialog sounds a lot like an ADHD brain's "inner voice" with less repetition. It has a childlike quality to it thst's kinda cute.
One thing I found interesting is that, unlike other LLM models, when I raised a doubt over something it said it didn't just blindly accept that you're right (even if you were wrong), but instead reviewed it's sources and conclusions and in the end stuck by it.
- It's trained by ChatGPT Distillation (IP legality now under review)
- It's restricting access now due to processing deficiencies, likely caused by trade restrictions
- It's power usage innovations are opensource, meaning all other models now have them + exorbitant processing meaning Deepseek may not maintain competitiveness
The final two paragraphs of this are the best take I have seen on AI and writing!! I edit for a living and my team has been getting skittish about AI’s influence in writing and editing, but the people who love to write will still write, and the people who love to edit will still edit, and the people who love to read things in their entirety instead of asking GPT or CliffsNotes for a summary will still read things in their entirety. Thank you for this!!
AI is ultimately tied to the military drone space, and why we should be very wary using it, regardless of its proposed benefits. Artificial Intelligence cannot create life - it assimilates life.
Personally I will not download any of these AI apps. I’m aware “Big Brother”, both domestic and international are listening and watching my every word,but I won’t sign up for any of it. I’m sticking with “BI” Biological Intelligence. I have four adult children who are brilliant and expert in their own professions. I turn to them when I need help or input.
It was an excellent reply however, his position on whether Deepseek used distillation to achieve that result was disappointing. Having an opinion matters. Good or bad.
I’ve been collecting data about DeepSeek. My experience has been very positive. I will be writing about it soon. I do not believe that LLM technology will in any way achieve AGI. At least DeepSeek knows this. Fifty+ years as an AI scientist. Dr. Randy Kaplan. (And by the way, let’s not forget Bill Gates accomplished quite a bit of corporate espionage and THEFT in the creation of Windows and it still was and is a terrible OS.)
I saw an article about this very topic this morning in Wired. That made be go and see what has been written about OpenAI’s security testing. I can’t find ONE article. If you have any references to this, please pass them on. What I do know is that I’ve been asking openAI to make their test plan available and they have not. I am not so confident that their record is any better than DeepSeek.
Never have tried an AI application ( not even spell check or garmmarly) and probably never will. I am 70, and know all too well that getting good info/intel these days is like having to mine for gold. The ore grades are degrading and so you have to dig deeper and sift through more rubble to get what you used to be able to pick up on the surface of a stream as a pure nugget.
Even worse, when I tried to post reviews of my lawyers In Elizabethtown and Louisville KY and the judges, Medical Boards, Hardin Memorial Hospital in KY and then later the NC Medical Board and Physician Health Program in NC, my Google, Facebook, and AVVO ( a lawyer feedback forum supposedly for clients to rate their lawyers) were shadow banned. You can type all the reviews you want and think you are posting them because you see them on your screen, but in reality these posts are never seen by the public. It is a double deception upon you the poster and the potential reader who might benefit from your post.
So basically, If and when I post anything on line, (especially with my browser Yahoo- a frequent blocker- which I have used since the 90's) even things that could theoretically be sourced through public records, I reluctantly question the futility of it all. My first question is always "Will this get censored?" Apparently China is already at it.
So against this background, AI seems pointless. You simply can't rely on it. That's why I got my HAM Radio Extra ratings etc.
Same age here. AI is pointless, for the reasons you mentioned and many more. The potential for abuse and manipulation is infinite and the more the day to day system embraces it, the more chaotic our lives will become.
The only solution I can see is to detach, dropout and disengage. Just one example- don't write reviews, don't challenge the system, as you've been doing. Instead, figure out how to go your own way. It will take some thought, some planning.
You can't let age keep you from doing so. Picture yourself as a 30 or 40 year old and go from there.
I still think the $5.6M number deserves more tire kicking. No company in China, as TikTok has shown, operates without government partnership and that a technology this transformative and game changing could catch the CCP off guard when they’ve historically been so involved in new tech, even down to the individual researcher level, just doesn’t pass muster. There’s been a long history of reporting lower than plausible defense budgets budgets (but to be fair why WOULD they tell us how much they’re spending on that?) and even changing the numbers with Olympic gymnasts’ ages and yet the media faithfully and dutifully reports whatever numbers they’re given each time. It’s not unfathomable that they would be willing to burn however much money it takes to create a narrative that would destabilize American markets.
All the paper shows is the breakdown and that the $5.6M only represents the final training round. That’s exactly my point - that statement is meaningless. It could still be true and the previous expenses could be $500B. It’s deliberately misleading.
I hear you. This stuff is new, and I’m getting tired and old. The foundation for this was laid way back and history is written by the winners. You have to search hard for the deeply buried truths.
I hope you stick to your writing. I have a food blog and I put a lot of work into the photos, recipe, text and the help I give and my photos are showing up on Amazon in AI recipes. I don't use AI because I want my readers to hear from me directly,. I just hope others feel the same.
I believe that writing prose with AI will ruin our ability to compose, much as spell-check has ruined our ability to spell. The second order effect is obvious when reading essays our other published texts. They are littered correctly spelled but incorrect words. A phenomenon that I’ve dubbed ‘death by spell check’
"The second-order effect is obvious when reading essays or other published texts. They are littered with correctly spelled but incorrect words." Copyright Copilot AI
I tried out DeepSeek for some research I was doing. It still produces errors and hallucinations like all LLMs. Personally, I didn't find it performed any better than Mistral, with the exception that it accessed more recent data. The "thinking" dialog sounds a lot like an ADHD brain's "inner voice" with less repetition. It has a childlike quality to it thst's kinda cute.
One thing I found interesting is that, unlike other LLM models, when I raised a doubt over something it said it didn't just blindly accept that you're right (even if you were wrong), but instead reviewed it's sources and conclusions and in the end stuck by it.
Some interesting points about Deepseek
- It's trained by ChatGPT Distillation (IP legality now under review)
- It's restricting access now due to processing deficiencies, likely caused by trade restrictions
- It's power usage innovations are opensource, meaning all other models now have them + exorbitant processing meaning Deepseek may not maintain competitiveness
the ability to do re-check instead of blindly being flaky does slightly pleases me
I love the way it explains stuff. I'm audhd, so, yeah...
Nice piece. Explained in a way an old man like me might learn more than 1/2 of what you conveyed.
The final two paragraphs of this are the best take I have seen on AI and writing!! I edit for a living and my team has been getting skittish about AI’s influence in writing and editing, but the people who love to write will still write, and the people who love to edit will still edit, and the people who love to read things in their entirety instead of asking GPT or CliffsNotes for a summary will still read things in their entirety. Thank you for this!!
AI is ultimately tied to the military drone space, and why we should be very wary using it, regardless of its proposed benefits. Artificial Intelligence cannot create life - it assimilates life.
https://romanshapoval.substack.com/p/drones-and-the-us-space-force
Personally I will not download any of these AI apps. I’m aware “Big Brother”, both domestic and international are listening and watching my every word,but I won’t sign up for any of it. I’m sticking with “BI” Biological Intelligence. I have four adult children who are brilliant and expert in their own professions. I turn to them when I need help or input.
Here here. I've got three kiddos. Hang around horses, not machines. People, not machines. Much better long term plan
The CEO of Anthropic wrote a blog post that clears up the deepseek hype. I highly recommend: https://darioamodei.com/on-deepseek-and-export-controls
It was an excellent reply however, his position on whether Deepseek used distillation to achieve that result was disappointing. Having an opinion matters. Good or bad.
The ceo of anthropic isn't an objective arbiter of hype 🤷♂️
I’ve been collecting data about DeepSeek. My experience has been very positive. I will be writing about it soon. I do not believe that LLM technology will in any way achieve AGI. At least DeepSeek knows this. Fifty+ years as an AI scientist. Dr. Randy Kaplan. (And by the way, let’s not forget Bill Gates accomplished quite a bit of corporate espionage and THEFT in the creation of Windows and it still was and is a terrible OS.)
MAC and Linux both suck LoL! I use all three and can't figure out which is worse :)
I hope you include information about its current security level. What I've been reading on that front is, put mildly, incredibly concerning.
I did some further looking around. Low and behold. Hmmm, "incredibly concerning," yes I would agree, when it comes to OpenAI. Here ya go.
https://www.forbes.com/sites/alexvakulov/2025/02/01/more-chatgpt-jailbreaks-are-evading-safeguards-on-sensitive-topics/
https://hbr.org/2023/04/the-new-risks-chatgpt-poses-to-cybersecurity
https://www.sentinelone.com/cybersecurity-101/data-and-ai/chatgpt-security-risks/
These are the first three that came up in a Google search.
I hope this satisfies your concerns about what I wrote.
I saw an article about this very topic this morning in Wired. That made be go and see what has been written about OpenAI’s security testing. I can’t find ONE article. If you have any references to this, please pass them on. What I do know is that I’ve been asking openAI to make their test plan available and they have not. I am not so confident that their record is any better than DeepSeek.
Time to 'DeepSeek" all this AI nonsense now!
Every 'human voice ' is a robot and another job gone just like "Self -Serve" in the grocery store.
The Corporate State is out to impoverish the masses world-wide. AI is just their latest scheme
It's strictly about control
AI's about control, and it may accomplish control through impoverishment. Or it may accomplish impoverishment through control. It doesn't care.
This was so interesting and valuable. Thank you for decoding this truly unimaginable world for the IT-challenged.
Never have tried an AI application ( not even spell check or garmmarly) and probably never will. I am 70, and know all too well that getting good info/intel these days is like having to mine for gold. The ore grades are degrading and so you have to dig deeper and sift through more rubble to get what you used to be able to pick up on the surface of a stream as a pure nugget.
Even worse, when I tried to post reviews of my lawyers In Elizabethtown and Louisville KY and the judges, Medical Boards, Hardin Memorial Hospital in KY and then later the NC Medical Board and Physician Health Program in NC, my Google, Facebook, and AVVO ( a lawyer feedback forum supposedly for clients to rate their lawyers) were shadow banned. You can type all the reviews you want and think you are posting them because you see them on your screen, but in reality these posts are never seen by the public. It is a double deception upon you the poster and the potential reader who might benefit from your post.
So basically, If and when I post anything on line, (especially with my browser Yahoo- a frequent blocker- which I have used since the 90's) even things that could theoretically be sourced through public records, I reluctantly question the futility of it all. My first question is always "Will this get censored?" Apparently China is already at it.
So against this background, AI seems pointless. You simply can't rely on it. That's why I got my HAM Radio Extra ratings etc.
Same age here. AI is pointless, for the reasons you mentioned and many more. The potential for abuse and manipulation is infinite and the more the day to day system embraces it, the more chaotic our lives will become.
The only solution I can see is to detach, dropout and disengage. Just one example- don't write reviews, don't challenge the system, as you've been doing. Instead, figure out how to go your own way. It will take some thought, some planning.
You can't let age keep you from doing so. Picture yourself as a 30 or 40 year old and go from there.
I still think the $5.6M number deserves more tire kicking. No company in China, as TikTok has shown, operates without government partnership and that a technology this transformative and game changing could catch the CCP off guard when they’ve historically been so involved in new tech, even down to the individual researcher level, just doesn’t pass muster. There’s been a long history of reporting lower than plausible defense budgets budgets (but to be fair why WOULD they tell us how much they’re spending on that?) and even changing the numbers with Olympic gymnasts’ ages and yet the media faithfully and dutifully reports whatever numbers they’re given each time. It’s not unfathomable that they would be willing to burn however much money it takes to create a narrative that would destabilize American markets.
Do actual research on what the $5.6m number entails - it's outlined in their paper!
All the paper shows is the breakdown and that the $5.6M only represents the final training round. That’s exactly my point - that statement is meaningless. It could still be true and the previous expenses could be $500B. It’s deliberately misleading.
it's not deliberately misleading lol it's assuming anyone interpreting it has basic comprehension skills
Americans are mad because a chinese company built an AI software that is so advance than anything else in the world and that too for peanuts.
I’ve a gut feeling that resistance is futile. But I’m going to behave in the opposite manner.
I have a feeling resistance is futile if you allow more than just a small amount of AI in your life. Try to work with it and it will consume you
I hear you. This stuff is new, and I’m getting tired and old. The foundation for this was laid way back and history is written by the winners. You have to search hard for the deeply buried truths.
I hope you stick to your writing. I have a food blog and I put a lot of work into the photos, recipe, text and the help I give and my photos are showing up on Amazon in AI recipes. I don't use AI because I want my readers to hear from me directly,. I just hope others feel the same.
My daughter has the same problem with her blog. No respect for copyrighted material.
I'll rally around AI when it experiences and understands love, grief, the agony of loss, and the intense thrill of redemption.
I believe that writing prose with AI will ruin our ability to compose, much as spell-check has ruined our ability to spell. The second order effect is obvious when reading essays our other published texts. They are littered correctly spelled but incorrect words. A phenomenon that I’ve dubbed ‘death by spell check’
Grammar or spelling errors? How about this-
"The second-order effect is obvious when reading essays or other published texts. They are littered with correctly spelled but incorrect words." Copyright Copilot AI
Kind of the point of the intentional errors