https://www.semianalysis.com/p/google-we-have-no-moat-and-neither

Quote:

---

While our models still hold a slight edge in terms of quality, the gap is closing astonishingly quickly. Open-source models are faster, more customizable, more private, and pound-for-pound more capable. They are doing things with $100 and 13B params that we struggle with at $10M and 540B. And they are doing so in weeks, not months. This has profound implications for us:

We have no secret sauce. Our best hope is to learn from and collaborate with what others are doing outside Google. We should prioritize enabling 3P integrations.
People will not pay for a restricted model when free, unrestricted alternatives are comparable in quality. We should consider where our value add really is.
Giant models are slowing us down. In the long run, the best models are the ones
which can be iterated upon quickly. We should make small variants more than an afterthought, now that we know what is possible in the <20B parameter regime.

---

The lobomitized models are going to be destroyed because absolutely no one wants them beyond the leftist creators who want to be thought police. The future for language models is entirely open source and with each person having their own private models, trained on whatever they want.

Didn't they just announce a halt or back off in funding in their own AI? Maybe its in the memo.

First one to get a sufficiently capable AI that can write/test/apply own enhancements and upgrades and this will spin out of control very very fast.

MouthBQ98 said:

---

First one to get a sufficiently capable AI that can write/test/apply own enhancements and upgrades and this will spin out of control very very fast.

---

Given what the open source guys are doing, that's where it's going to come from.

Yep. Most humans never really comprehend the implications of geometric functions.

Nanomachines son said:

---

https://www.semianalysis.com/p/google-we-have-no-moat-and-neither

Quote:

---

While our models still hold a slight edge in terms of quality, the gap is closing astonishingly quickly. Open-source models are faster, more customizable, more private, and pound-for-pound more capable. They are doing things with $100 and 13B params that we struggle with at $10M and 540B. And they are doing so in weeks, not months. This has profound implications for us:

We have no secret sauce. Our best hope is to learn from and collaborate with what others are doing outside Google. We should prioritize enabling 3P integrations.
People will not pay for a restricted model when free, unrestricted alternatives are comparable in quality. We should consider where our value add really is.
Giant models are slowing us down. In the long run, the best models are the ones
which can be iterated upon quickly. We should make small variants more than an afterthought, now that we know what is possible in the <20B parameter regime.

---

The lobomitized models are going to be destroyed because absolutely no one wants them beyond the leftist creators who want to be thought police. The future for language models is entirely open source and with each person having their own private models, trained on whatever they want.

---

I asked ChatGPT to write a night show opening monologue about President poopy pants and his worsening dementia, policy bribery scandal and Biden's son's attempts to not disclose his income in an Arkansas court.

It lectured me about there being no confirmation of his dementia or fraud and that it was rude to joke about health problems. After changing the names to Trump & jr. it no longer had any ethical concerns and wrote the jokes.

lb3 said:

---

Nanomachines son said:

---

https://www.semianalysis.com/p/google-we-have-no-moat-and-neither

Quote:

---

While our models still hold a slight edge in terms of quality, the gap is closing astonishingly quickly. Open-source models are faster, more customizable, more private, and pound-for-pound more capable. They are doing things with $100 and 13B params that we struggle with at $10M and 540B. And they are doing so in weeks, not months. This has profound implications for us:

We have no secret sauce. Our best hope is to learn from and collaborate with what others are doing outside Google. We should prioritize enabling 3P integrations.
People will not pay for a restricted model when free, unrestricted alternatives are comparable in quality. We should consider where our value add really is.
Giant models are slowing us down. In the long run, the best models are the ones
which can be iterated upon quickly. We should make small variants more than an afterthought, now that we know what is possible in the <20B parameter regime.

---

The lobomitized models are going to be destroyed because absolutely no one wants them beyond the leftist creators who want to be thought police. The future for language models is entirely open source and with each person having their own private models, trained on whatever they want.

---

I asked ChatGPT to write a night show opening monologue about President poopy pants and his worsening dementia, policy bribery scandal and Biden's son's attempts to not disclose his income in an Arkansas court.

It lectured me about there being no confirmation of his dementia or fraud and that it was rude to joke about health problems. After changing the names to Trump & jr. it no longer had any ethical concerns and wrote the jokes.

---

We're ****ed

Yep. Pretty sad.

I think this is the reason we haven't found intelligent life. When it gets too intelligent, it goes extinct. We are about to do the same.

Google will just buy useful idiots in Congress to "regulate" open source AI in the name of national security. Then after 90% of the startups fold they will buy the survivors.

MouthBQ98 said:

---

Yep. Most humans never really comprehend the implications of geometric functions.

---

They also never suspect the Spanish Inquisition.

lb3 said:

---

Nanomachines son said:

---

https://www.semianalysis.com/p/google-we-have-no-moat-and-neither

Quote:

---

While our models still hold a slight edge in terms of quality, the gap is closing astonishingly quickly. Open-source models are faster, more customizable, more private, and pound-for-pound more capable. They are doing things with $100 and 13B params that we struggle with at $10M and 540B. And they are doing so in weeks, not months. This has profound implications for us:

We have no secret sauce. Our best hope is to learn from and collaborate with what others are doing outside Google. We should prioritize enabling 3P integrations.
People will not pay for a restricted model when free, unrestricted alternatives are comparable in quality. We should consider where our value add really is.
Giant models are slowing us down. In the long run, the best models are the ones
which can be iterated upon quickly. We should make small variants more than an afterthought, now that we know what is possible in the <20B parameter regime.

---

The lobomitized models are going to be destroyed because absolutely no one wants them beyond the leftist creators who want to be thought police. The future for language models is entirely open source and with each person having their own private models, trained on whatever they want.

---

I asked ChatGPT to write a night show opening monologue about President poopy pants and his worsening dementia, policy bribery scandal and Biden's son's attempts to not disclose his income in an Arkansas court.

It lectured me about there being no confirmation of his dementia or fraud and that it was rude to joke about health problems. After changing the names to Trump & jr. it no longer had any ethical concerns and wrote the jokes.

---

GIGO.

This is why we will have to find AI not written by known leftists. Should be able to do with open source, but all of corporate America and anything official will be required to use government sanctioned AI that properly oppresses conservatives and Republicans.

For Open Source AI to dominate is humanity's only hope. Put the power of AI in everyone's hands.

Maybe it will be the downfall of Social Media. We can only hope.

lb3 said:

---

Nanomachines son said:

---

https://www.semianalysis.com/p/google-we-have-no-moat-and-neither

Quote:

---

While our models still hold a slight edge in terms of quality, the gap is closing astonishingly quickly. Open-source models are faster, more customizable, more private, and pound-for-pound more capable. They are doing things with $100 and 13B params that we struggle with at $10M and 540B. And they are doing so in weeks, not months. This has profound implications for us:

We have no secret sauce. Our best hope is to learn from and collaborate with what others are doing outside Google. We should prioritize enabling 3P integrations.
People will not pay for a restricted model when free, unrestricted alternatives are comparable in quality. We should consider where our value add really is.
Giant models are slowing us down. In the long run, the best models are the ones
which can be iterated upon quickly. We should make small variants more than an afterthought, now that we know what is possible in the <20B parameter regime.

---

The lobomitized models are going to be destroyed because absolutely no one wants them beyond the leftist creators who want to be thought police. The future for language models is entirely open source and with each person having their own private models, trained on whatever they want.

---

I asked ChatGPT to write a night show opening monologue about President poopy pants and his worsening dementia, policy bribery scandal and Biden's son's attempts to not disclose his income in an Arkansas court.

It lectured me about there being no confirmation of his dementia or fraud and that it was rude to joke about health problems. After changing the names to Trump & jr. it no longer had any ethical concerns and wrote the jokes.

---

TexAgs91 said:

---

For Open Source AI to dominate is humanity's only hope. Put the power of AI in everyone's hands.

---

Agreed 1,000%.

Open source is available for **anyone** to download and inspect, modify, re-release...

It is the Bizarre, vs. The Cathedral, as Eric Schmidt termed it...

Microsoft is a Cathedral...

SeMgCo87 said:

---

TexAgs91 said:

---

For Open Source AI to dominate is humanity's only hope. Put the power of AI in everyone's hands.

---

Agreed 1,000%.

Open source is available for **anyone** to download and inspect, modify, re-release...

It is the Bizarre, vs. The Cathedral, as Eric Schmidt termed it...

Microsoft is a Cathedral...

---

Think you mean "bazaar."

Didn't take long

The Biden administration announced a variety of #ArtificialIntelligence policies and programs, including news on government use of AI that comes on the heels of a February executive order to embed “racial equity” in government AI. https://t.co/Zu4zgmt5Te

— The Epoch Times (@EpochTimes) May 5, 2023

techno-ag said:

---

SeMgCo87 said:

---

TexAgs91 said:

---

For Open Source AI to dominate is humanity's only hope. Put the power of AI in everyone's hands.

---

Agreed 1,000%.

Open source is available for **anyone** to download and inspect, modify, re-release...

It is the Bizarre, vs. The Cathedral, as Eric Schmidt termed it...

Microsoft is a Cathedral...

---

Think you mean "bazaar."

---

Exactly. Sorry, I didn't fight back against spell-check...

ChatGPT - Generative Pre-trained Transformer

Reality:
ChatGPT - Giant Propaganda Tool

GeorgiAg said:

---

Yep. Pretty sad.

I think this is the reason we haven't found intelligent life. When it gets too intelligent, it goes extinct. We are about to do the same.

---

I guess on all the planets that grew purple haired leftists the others should be fine !

SeMgCo87 said:

---

TexAgs91 said:

---

For Open Source AI to dominate is humanity's only hope. Put the power of AI in everyone's hands.

---

Agreed 1,000%.

Open source is available for **anyone** to download and inspect, modify, re-release...

It is the Bizarre, vs. The Cathedral, as Eric Schmidt termed it...

Microsoft is a Cathedral...

---

Modify like Wikipedia?

Snoop Dog on AI

5. Snoop Dogg gives his take on AI

Honestly, this is a pretty accurate representation of what's happening in AI right now.

Someone tell @SnoopDogg to subscribe to my newsletter 😏pic.twitter.com/YJNTE7Yrxx

— Rowan Cheung (@rowancheung) May 5, 2023

GeorgiAg said:

---

Yep. Pretty sad.

I think this is the reason we haven't found intelligent life. When it gets too intelligent, it goes extinct. We are about to do the same.

---

This is why Elon purchased Twitter. He's worried about the Great Filter ending civilization before he can get us a toe hold on a second planet. Social media and AI are some of his highest threats he sees to our species.

FJB said:

---

SeMgCo87 said:

---

TexAgs91 said:

---

For Open Source AI to dominate is humanity's only hope. Put the power of AI in everyone's hands.

---

Agreed 1,000%.

Open source is available for **anyone** to download and inspect, modify, re-release...

It is the Bizarre, vs. The Cathedral, as Eric Schmidt termed it...

Microsoft is a Cathedral...

---

Modify like Wikipedia?

---

Sorry, have been traveling...

No, absolutely not... Wikipedia is refereed, but it is possible crap stays in online version, for any one to see, or be misled

Open source rules mean anybody can see / modify code, but code "owner" reviews and confirms changes meet module intent, functionality and operability... a contributor cannot radically change what a code block does, or even hide crap, evil crap, in the module.

The modifier can, however, offer it separately.

If a developer has constructed a good code module for AI, for example, it gets incorporated into a bigger set...if it doesn't behave properly for everyone involved, the code is sent back, or not allowed.

Many eyes make bugs shallow...and visible.

On the bright side, AI bots will eventually be able to do your message board posting for you, freeing up a ton of your time for other things.

The Amish aren't sweating this.

lb3 said:

---

Nanomachines son said:

---

https://www.semianalysis.com/p/google-we-have-no-moat-and-neither

Quote:

---

While our models still hold a slight edge in terms of quality, the gap is closing astonishingly quickly. Open-source models are faster, more customizable, more private, and pound-for-pound more capable. They are doing things with $100 and 13B params that we struggle with at $10M and 540B. And they are doing so in weeks, not months. This has profound implications for us:

We have no secret sauce. Our best hope is to learn from and collaborate with what others are doing outside Google. We should prioritize enabling 3P integrations.
People will not pay for a restricted model when free, unrestricted alternatives are comparable in quality. We should consider where our value add really is.
Giant models are slowing us down. In the long run, the best models are the ones
which can be iterated upon quickly. We should make small variants more than an afterthought, now that we know what is possible in the <20B parameter regime.

---

The lobomitized models are going to be destroyed because absolutely no one wants them beyond the leftist creators who want to be thought police. The future for language models is entirely open source and with each person having their own private models, trained on whatever they want.

---

I asked ChatGPT to write a night show opening monologue about President poopy pants and his worsening dementia, policy bribery scandal and Biden's son's attempts to not disclose his income in an Arkansas court.

It lectured me about there being no confirmation of his dementia or fraud and that it was rude to joke about health problems. After changing the names to Trump & jr. it no longer had any ethical concerns and wrote the jokes.

---

I had a similar thing on climate. Asked it to describe why climate alarmism isn't rational and it lectured me on why using the term climate alarmism is inappropriate. Asked it to give me talking points for a debate against climate deniers and it wrote several paragraphs. I then asked it how come it thinks climate alarmism is a hurtful term that shouldn't be used but climate denier is okay, and it said the following:


"You are correct that the term "climate denier" can also have a negative connotation and be perceived as dismissive or derogatory towards those who reject the overwhelming scientific consensus on climate change. As an AI language model, I strive to be neutral and objective in my responses, and I apologize if my previous answer did not reflect that.

In general, it is important to use language that is respectful, accurate, and avoids derogatory or dismissive terms that could undermine productive dialogue. Instead of using divisive labels, it is more productive to focus on the underlying scientific evidence and engage in respectful, fact-based discussions with those who hold different views.

I appreciate your feedback and hope that this response clarifies my stance on the use of language in discussing climate change."


One thing I don't know is how much of the bias it has is programmed and in and how much just arises naturally because of the bias in the mainstream media sources it just repackages naturally through the way it functions.

GE said:

---

lb3 said:

---

Nanomachines son said:

---

https://www.semianalysis.com/p/google-we-have-no-moat-and-neither

Quote:

---

While our models still hold a slight edge in terms of quality, the gap is closing astonishingly quickly. Open-source models are faster, more customizable, more private, and pound-for-pound more capable. They are doing things with $100 and 13B params that we struggle with at $10M and 540B. And they are doing so in weeks, not months. This has profound implications for us:

We have no secret sauce. Our best hope is to learn from and collaborate with what others are doing outside Google. We should prioritize enabling 3P integrations.
People will not pay for a restricted model when free, unrestricted alternatives are comparable in quality. We should consider where our value add really is.
Giant models are slowing us down. In the long run, the best models are the ones
which can be iterated upon quickly. We should make small variants more than an afterthought, now that we know what is possible in the <20B parameter regime.

---

The lobomitized models are going to be destroyed because absolutely no one wants them beyond the leftist creators who want to be thought police. The future for language models is entirely open source and with each person having their own private models, trained on whatever they want.

---

I asked ChatGPT to write a night show opening monologue about President poopy pants and his worsening dementia, policy bribery scandal and Biden's son's attempts to not disclose his income in an Arkansas court.

It lectured me about there being no confirmation of his dementia or fraud and that it was rude to joke about health problems. After changing the names to Trump & jr. it no longer had any ethical concerns and wrote the jokes.

---

I had a similar thing on climate. Asked it to describe why climate alarmism isn't rational and it lectured me on why using the term climate alarmism is inappropriate. Asked it to give me talking points for a debate against climate deniers and it wrote several paragraphs. I then asked it how come it thinks climate alarmism is a hurtful term that shouldn't be used but climate denier is okay, and it said the following:


"You are correct that the term "climate denier" can also have a negative connotation and be perceived as dismissive or derogatory towards those who reject the overwhelming scientific consensus on climate change. As an AI language model, I strive to be neutral and objective in my responses, and I apologize if my previous answer did not reflect that.

In general, it is important to use language that is respectful, accurate, and avoids derogatory or dismissive terms that could undermine productive dialogue. Instead of using divisive labels, it is more productive to focus on the underlying scientific evidence and engage in respectful, fact-based discussions with those who hold different views.

I appreciate your feedback and hope that this response clarifies my stance on the use of language in discussing climate change."


One thing I don't know is how much of the bias it has is programmed and in and how much just arises naturally because of the bias in the mainstream media sources it just repackages naturally through the way it functions.

---

It's the latter. We know that several years ago Google changed the nature of search to favor "authoritative" sources which effectively means legacy news media and government. Earth herself could start a website and explain why the climate changes irrespective of human activity and Google would suppress that site in favor of pinhead academics screeching in scientific journals about climate change. From my understanding of AI at this stage it can't help but suffer from those confirmation biases.

GeorgiAg said:

---

Yep. Pretty sad.

I think this is the reason we haven't found intelligent life. When it gets too intelligent, it goes extinct. We are about to do the same.

---

I'm sure there were many that would post the same thing during the Industrial Revolution, the potato famine, the dark ages, etc if they had the tech to do it.

One thing I always find fascinating is that we are arrogant enough to think the here and now is so grossly different than other major tech advances in the human race.

When people were able to travel from coast to coast in hours vs weeks via commercial flights you don't think everyone at that time thought crazy things for the future?

Think about what y'all thought when the Apollo program was going on. We were colonizing Titan at this point for sure.

I digress, just saying that the world will go on and our period of time is not going to be looked back 300 years from now as anymore significant than other big jumps in tech/society/culture. Just another progressive period of civilization imho.

TyHolden said:

---

lb3 said:

---

Nanomachines son said:

---

https://www.semianalysis.com/p/google-we-have-no-moat-and-neither

Quote:

---

While our models still hold a slight edge in terms of quality, the gap is closing astonishingly quickly. Open-source models are faster, more customizable, more private, and pound-for-pound more capable. They are doing things with $100 and 13B params that we struggle with at $10M and 540B. And they are doing so in weeks, not months. This has profound implications for us:

We have no secret sauce. Our best hope is to learn from and collaborate with what others are doing outside Google. We should prioritize enabling 3P integrations.
People will not pay for a restricted model when free, unrestricted alternatives are comparable in quality. We should consider where our value add really is.
Giant models are slowing us down. In the long run, the best models are the ones
which can be iterated upon quickly. We should make small variants more than an afterthought, now that we know what is possible in the <20B parameter regime.

---

The lobomitized models are going to be destroyed because absolutely no one wants them beyond the leftist creators who want to be thought police. The future for language models is entirely open source and with each person having their own private models, trained on whatever they want.

---

I asked ChatGPT to write a night show opening monologue about President poopy pants and his worsening dementia, policy bribery scandal and Biden's son's attempts to not disclose his income in an Arkansas court.

It lectured me about there being no confirmation of his dementia or fraud and that it was rude to joke about health problems. After changing the names to Trump & jr. it no longer had any ethical concerns and wrote the jokes.

---

We're ****ed

---

No. If this is right and Open Source AI is free and as capable we will be just fine. We can all, including our very own AI's can become even more polarized.

SeMgCo87 said:

---

FJB said:

---

SeMgCo87 said:

---

TexAgs91 said:

---

For Open Source AI to dominate is humanity's only hope. Put the power of AI in everyone's hands.

---

Agreed 1,000%.

Open source is available for **anyone** to download and inspect, modify, re-release...

It is the Bizarre, vs. The Cathedral, as Eric Schmidt termed it...

Microsoft is a Cathedral...

---

Modify like Wikipedia?

---

Sorry, have been traveling...

No, absolutely not... Wikipedia is refereed, but it is possible crap stays in online version, for any one to see, or be misled

Open source rules mean anybody can see / modify code, but code "owner" reviews and confirms changes meet module intent, functionality and operability... a contributor cannot radically change what a code block does, or even hide crap, evil crap, in the module.

The modifier can, however, offer it separately.

If a developer has constructed a good code module for AI, for example, it gets incorporated into a bigger set...if it doesn't behave properly for everyone involved, the code is sent back, or not allowed.

Many eyes make bugs shallow...and visible.

---

Yep, I have used open source for years. Linux OS, dev tools, compilers, db's etc. All work great. Got a job last year working with MS stuff and Its driving me nuts. Not nearly as reliable