The thesis that LLM development has already peaked - do you think it's plausible

[unwieldy]

Quite a number of IT-literate people have pointed out that LLM development has realistically already peaked, because it absorbs, synthesizes and plagiarizes human-created content. Considering most human content has already been harvested, and the rate of this content expansion is slow, many feel that LLMs won't go much farther from here. There are also studies indicating that people who use LLMs a lot are eroding their critical thinking abilities, so logically the rate of human progress that can be plagiarized in the first place may actually inverse.

Your thoughts on the matter?

[Hulkotron]

Will wait to hear Lyle McDonald's thoughts before I fully erect my opinion on the matter.

[Raymondo]

Not sure human-created content can truly be exhausted since new content is generated every day and at ever increasing quantities, plus it usually mixed with synthetic data anyway

[unwieldy]

Not sure human-created content can truly be exhausted since new content is generated every day and at ever increasing quantities, plus it usually mixed with synthetic data anyway

I also considered that millions of corporate enrolled computers are harvesting everything highly skilled professionals are doing everyday. But on the other hand, if people start relying more and more on LLMs for problem solving at their job, this could affect what can be extracted from the human brain.

[Raymondo]

I also considered that millions of corporate enrolled computers are harvesting everything highly skilled professionals are doing everyday. But on the other hand, if people start relying more and more on LLMs for problem solving at their job, this could affect what can be extracted from the human brain.

I think the vast majority of human generated content is not accessible to LLMs.

Take code for instance, CoPilot is not allowed to use private repositories for training. Yet the vast majority of high quality code will be locked behind intellectual property laws, not publicly available. I don't know if it's the same with other types of content, but I imagine literature could be the same, if someone owns the copyright an LLM cannot use it in its model.

[delon]

Getbig is currently being scraped by at least a few thousand bots (6,000 'Guests'  ) so as long as we continue to produce our standard high level of insightful content the trajectory of collective human knowledge will remain positive

This is our gift to the world, what we do here matters

[unwieldy]

I think the vast majority of human generated content is not accessible to LLMs.

Take code for instance, CoPilot is not allowed to use private repositories for training. Yet the vast majority of high quality code will be locked behind intellectual property laws, not publicly available. I don't know if it's the same with other types of content, but I imagine literature could be the same, if someone owns the copyright an LLM cannot use it in its model.

I'm glad you brought up CoPilot, this was one thing I was thinking about in the post you quoted. As so many knowledge workers are using windows 11 which supposedly has copilot integrated into everything, I was wondering about that. Could you give some additional insights on that? Thanks.

[unwieldy]

Getbig is currently being scraped by at least a few thousand bots (6,000 'Guests'  ) so as long as we continue to produce our standard high level of insightful content the trajectory of collective human knowledge will remain positive

This is our gift to the world, what we do here matters

I considered investigating in scalable misdirection content generation with totally wrong info presented as rock solid.

[Raymondo]

I'm glad you brought up CoPilot, this was one thing I was thinking about in the post you quoted. As so many knowledge workers are using windows 11 which supposedly has copilot integrated into everything, I was wondering about that. Could you give some additional insights on that? Thanks.

If it's integrated into Win 11 (not a Win user so don't know) , I think it's safe to say they won't use it on the content users create, but may use it to train on usage patterns/telemetry etc.

I know that in terms of software development LLMs have changed the landscape significantly in the last six months and will continue to do so. They are just too good to be true.

It is difficult to predict where this will lead and there are heated discussions all the time in places like hackernews, etc. A consensus is that at least for the next few years the junior developer role will largely disappear. It is already difficult enough to get into entry level positions and LLMs will make it nigh impossible. But it is not sustainable long term. Seniors are still needed and will be needed for a long time, and with attrition/retirement rates there might be a scarcity of developer roles simply because there will not have been enough juniors gaining experience in prior years.

[unwieldy]

If it's integrated into Win 11 (not a Win user so don't know) , I think it's safe to say they won't use it on the content users create, but may use it to train on usage patterns/telemetry etc.

I know that in terms of software development LLMs have changed the landscape significantly in the last six months and will continue to do so. They are just too good to be true.

It is difficult to predict where this will lead and there are heated discussions all the time in places like hackernews, etc. A consensus is that at least for the next few years the junior developer role will largely disappear. It is already difficult enough to get into entry level positions and LLMs will make it nigh impossible. But it is not sustainable long term. Seniors are still needed and will be needed for a long time, and with attrition/retirement rates there might be a scarcity of developer roles simply because there will not have been enough juniors gaining experience in prior years.

That makes one wonder what will happen when we run out of seniors, from what said it seems like that pool will be depleted as no new juniors are starting out on the path to senior.

[unwieldy]

If it's integrated into Win 11 (not a Win user so don't know) , I think it's safe to say they won't use it on the content users create, but may use it to train on usage patterns/telemetry etc.

I know that in terms of software development LLMs have changed the landscape significantly in the last six months and will continue to do so. They are just too good to be true.

It is difficult to predict where this will lead and there are heated discussions all the time in places like hackernews, etc. A consensus is that at least for the next few years the junior developer role will largely disappear. It is already difficult enough to get into entry level positions and LLMs will make it nigh impossible. But it is not sustainable long term. Seniors are still needed and will be needed for a long time, and with attrition/retirement rates there might be a scarcity of developer roles simply because there will not have been enough juniors gaining experience in prior years.

That makes one wonder what will happen when we run out of seniors, from what said it seems like that pool will be depleted as no new juniors are starting out on the path to senior.

[unwieldy]

If it's integrated into Win 11 (not a Win user so don't know) , I think it's safe to say they won't use it on the content users create, but may use it to train on usage patterns/telemetry etc.

I know that in terms of software development LLMs have changed the landscape significantly in the last six months and will continue to do so. They are just too good to be true.

It is difficult to predict where this will lead and there are heated discussions all the time in places like hackernews, etc. A consensus is that at least for the next few years the junior developer role will largely disappear. It is already difficult enough to get into entry level positions and LLMs will make it nigh impossible. But it is not sustainable long term. Seniors are still needed and will be needed for a long time, and with attrition/retirement rates there might be a scarcity of developer roles simply because there will not have been enough juniors gaining experience in prior years.

That makes one wonder what will happen when we run out of seniors, from what said it seems like that pool will be depleted as no new juniors are starting out on the path to senior.

[unwieldy]

If it's integrated into Win 11 (not a Win user so don't know) , I think it's safe to say they won't use it on the content users create, but may use it to train on usage patterns/telemetry etc.

I know that in terms of software development LLMs have changed the landscape significantly in the last six months and will continue to do so. They are just too good to be true.

It is difficult to predict where this will lead and there are heated discussions all the time in places like hackernews, etc. A consensus is that at least for the next few years the junior developer role will largely disappear. It is already difficult enough to get into entry level positions and LLMs will make it nigh impossible. But it is not sustainable long term. Seniors are still needed and will be needed for a long time, and with attrition/retirement rates there might be a scarcity of developer roles simply because there will not have been enough juniors gaining experience in prior years.

That makes one wonder what will happen when we run out of seniors, from what said it seems like that pool will be depleted as no new juniors are starting out on the path to senior.

[Van_Bilderass]

I only saw some headlines a while back but what's become of the new Chatgpt controversy, is the new release any good?

[Palumboism]

I only saw some headlines a while back but what's become of the new Chatgpt controversy, is the new release any good?

Psychiatrist went to OpenAI, the company that owns Chatgpt and asked them to tone down the sycophancy because it was affecting their patients.  When ChatGpt came out with GPT-5 they go rid of GPT-4O which had a friendlier personality.  There was immediate revolt and they brought back GPT-4O as well. 

Gpt-5 was made for business, in particular companies that wanted to use it for programming.  Anthropic dominated the corporate programming market and GPT-5 was designed to gain market share.  The average user doesn't notice where they made the improvement and didn't like the cold personality of the new model.

I total, Gpt-5 is viewed by all to be a huge flop and the CEO of the company agrees with that.  They are already talking about GPT-6.

The fact that GPT-5 was a flop has people talking about LLM reaching their peak.




 

[Van_Bilderass]

Psychiatrist went to OpenAI the company that owns Chatgpt and asked them to tone down the sycophancy because it was affecting their patients.  When ChatGpt came out with GPT-5 they go rid of GPT-4O which had a friendlier personality.  There was immediate revolt and they brought back GPT-4O as well. 

Gpt-5 was made for business, in particular companies that wanted to use it for programming.  Anthropic dominated the corporate programming market and GPT-5 was designed to gain market share.  The average user doesn't notice where they made the improvement and didn't like the cold personality of the new model.

I total, Gpt-5 is viewed by all to be a huge flop and the CEO of the company agrees with that.  They are already talking about GPT-6.

The fact that GPT-5 was a flop has people talking about LLM reaching their peak.

Thanks.

Has or will the flop affect the company's valuation?

[Methyl m1ke]

Theres certainly logic to the idea. Basically AI is not actually thinking, its not creating, it is exactly what we wanted it to be, and hopefully will not become what we fear it could be.

[SweetDaddySiki]

But can LLM give you the natty or not gains that you need? ChatGPT told me that I need to eat a lot of protein and focus on squats, deadlifts, bench presses, standing overhead presses, and rows 3 times a week. I get the same info when I google or search YouTube. 

[Palumboism]

Quite a number of IT-literate people have pointed out that LLM development has realistically already peaked, because it absorbs, synthesizes and plagiarizes human-created content. Considering most human content has already been harvested, and the rate of this content expansion is slow, many feel that LLMs won't go much farther from here. There are also studies indicating that people who use LLMs a lot are eroding their critical thinking abilities, so logically the rate of human progress that can be plagiarized in the first place may actually inverse.

Your thoughts on the matter?

I would say LLM are where the internet was in 1999.  Google had just moved out of the garage, but wouldn't go public for five more years.  Three of the four biggest LLM have not gone public yet, which has created some of the biggest unicorns in history.  Anthropic is currently doing a capital raise based on a valuation of $170 Billion.  OpenAI is valued at over $300 billion. 


There are at least five tech companies with budgets of over $100 billion for compute next year.  Nvidia is caught in the middle of this gold rush selling shovels.

To quote Anthropic CEO, if you aren't making improvements by adding more compute and data, you're doing something wrong.  That statement is guiding the whole industry right now. 

Saying LLM have peaked in 2025 is the equivalent to saying the internet peaked in 1999. 

[Palumboism]

Thanks.

Has or will the flop affect the company's valuation?

If GPT-5 was as successful as GPT-4 it could have easily added $100 billion to the valuation.  In it's current form it added nothing.

For commercial users GPT-5 is a huge success and will possibly allow them to surpass Anthropic.

The other problem OpenAI has is they are by far the biggest, so if they make a miss step, everyone is watching and making predictions about the whole industry.