[Hidden-tech] Sharing LLM Insights:  Google’s NoteBook LM, SEO, Designers' Positive Career Implications

Rob Laporte rob at 2disc.com
Thu Jan 23 14:42:39 UTC 2025


Regarding both LLMs and open-source sharing in an association, this 8.5-minute Youtube shows truly stunning results: https://youtu.be/yEkAdyoZnj0?si=JGQ5rq_7ARsvY2bR. Gemini Pro summary below. Upshot: China's balanced management of communal capitalism enabled it to beat America's best LLMs using to the tune of less than 1% the cost. Sharing core intellectual capital and knowledge lifts everyone.

Gemini Summary:


The video is about how China is using an open-source approach to rapidly advance in AI and other technologies. Here’s a summary with the key points:

China's Open Source Advantage:

The video describes China's economy as "open source," where innovations are quickly shared and adapted across industries. This is facilitated by industrial clusters, similar to Silicon Valley but on a much larger scale.


The open-source approach allows Chinese companies to overcome limitations, such as semiconductor bans, and rapidly develop advanced AI models.


The video highlights the case of Deep Seek, a Chinese startup that built an AI model surpassing those from Facebook and OpenAI in just two months with a budget of less than $6 million.


In comparison, developing a similar model in the US would typically require 16,000 GPUs and significantly more time and resources. Deep Seek achieved this with only 2,48 GPUs and 11 times fewer GPU hours.


The video emphasizes that China's open-source strategy enables faster innovation and lower development costs. By opening their models to a vast network of developers, they accelerate progress and create a "rising tide" that benefits the entire ecosystem.



Impact on the US:

Eric Schmidt, former Google CEO and advisor on US AI policy, initially believed that the semiconductor export restrictions would hinder China's progress, giving the US a lead of two to three years.


However, he later admitted that he was wrong, as Chinese companies like Alibaba and Tencent quickly surpassed US companies in AI development using open-source models and older chips.


The video concludes by suggesting that understanding China's approach is crucial for US policymakers and that the focus should be on comprehending China's strategy rather than solely relying on technological superiority.



Key Numbers:

Time: Deep Seek developed its AI model in two months, while comparable models in the US take significantly longer.


Cost: Deep Seek spent less than $6 million, whereas similar models in the US would require vastly larger budgets.


GPUs: Deep Seek used 2,48 GPUs, compared to the estimated 16,000 GPUs needed for a comparable model in the US.


GPU Hours: Deep Seek used 11 times fewer GPU hours than what was previously thought necessary for such a model. 







Best Regards,



Rob Laporte

CEO  |  R&D Manager

DISC - Making Websites Make Money
mailto:Rob at 2disc.com, 413-584-6500

https://www.2disc.com



NOTE: Emails can be blocked by spam filters throughout the web. If you don’t get a reply within an expected span of time, please call.








---- On Thu, 23 Jan 2025 09:16:20 -0500 Rob Laporte <rob at 2disc.com> wrote ---



Hi Fungai,

Great tips, and thanks. You may know that Notebook LM enables one-click adding of a conversation or other output as a note to add to the Knowledge Base. Would be good if that feature was in ChatGPT. 

Regarding reminding LLMs of key info, I've noticed that one can consume the context window or token limits without knowing, which increases LLMs' forgetting. I once uploaded a single web page as raw code, not a pdf of content only, and the extensive code burnt almost all context in one shot. I saw that in Claude, which helpfully shows percent of context window used by both the Knowledge Base of uploaded files and by a current thread that may have its own uploads. I made standard procedures for dealing with this when one is not shown percent filled. The future association I envision would share pointers like this, helping lift all boats. 

Speculation: I've seen evidence that one's available context window may shrink if the LLM is currently being heavily used worldwide. I suspect that LLMs will have to charge more per month to offer power that is now given as a loss-leader. On the other hand, great advances are happening now to deliver more power per C/GPU so that micro-, small-, and medium-sized firms may not be short-changed relative to big business. 


Best Regards,



Rob Laporte

CEO  |  R&D Manager

DISC - Making Websites Make Money
mailto:Rob at 2disc.com, 413-584-6500

https://www.2disc.com



NOTE: Emails can be blocked by spam filters throughout the web. If you don’t get a reply within an expected span of time, please call.








---- On Thu, 23 Jan 2025 00:41:03 -0500 Fungai Tichawangana [Artist Dynamix] via Hidden-discuss <mailto:hidden-discuss at lists.hidden-tech.net> wrote ---












You're right, Rob. Notebook LM is amazing. I've found it great at giving summaries of long PDF reports.

And the podcasts are crazy good. 



My tip is related to Chat GPT. It is an impressive brainstorming partner- when you set the contextual groundwork. Over the course of a week or so, I'll share with it bits and pieces of info related to a project I'm working on; ideas, links to resources, related podcast transcripts, etc. Then one day, we sit down; GPT and I, and have a brainstorming session on that project. It is mind-blowing.



I've also found that naming things, helps Chat GPT get back on track when it's confused. For instance, at the end of a brainstorming session, I'll say, "Let's put everything we discussed today in an imaginary folder called 'Brainstorm 1x'.



If the conversation goes on over several days, I can always bring up the things we first discussed, by saying 'Remind me what the main takeaways from Brainstorm 1x were.'





Open skies,

Fungai 

Your Web Guy






---



Inc413, our flagship project brings small businesses 
in Western Massachusetts together. https://inc413.com/.


Meet with me: https://meet.artistdynamix.com/fungai.



---









Fungai 'Your Web Guy'
Tichawangana

Head Web Developer 
 & Online Visibility Strategist

mailto:fungai at artistdynamix.com

http://www.artistdynamix.comDigital Media Solutions 
for Creative Entrepreneurs 


https://www.instagram.com/artistdynamixhttps://www.facebook.com/artistdynamixhttps://www.twitter.com/artist_dynamix










Take the guesswork out of website performance—
run the 5-minute https://artistdynamix.com/jackchecker/?utm_source=email-signature&utm_medium=email&utm_campaign=Jackchecker+Email+Footer+Dec+2024 today!
























On Wed, 22 Jan 2025 at 16:10, Rob Laporte via Hidden-discuss <mailto:hidden-discuss at lists.hidden-tech.net> wrote:




_______________________________________________

Hidden-discuss mailing list - home page: http://www.hidden-tech.net 

mailto:Hidden-discuss at lists.hidden-tech.net 

 

You are receiving this because you are on the Hidden-Tech Discussion list. 

If you would like to change your list preferences, Go to the Members 

page on the Hidden Tech Web site. 

http://www.hidden-tech.net/members 




Hi All, 

I experimented with Google’s NoteBook LM, in which one can upload topic-specific files for summaries, nuanced insights, and finding needles in the haystack. I fed it over 30 top articles on how LLM-related tech in search engines has profoundly and probably permanently altered what one should do in SEO. NoteBook LM can generate a podcast-style conversation featuring male and female AI personas. With careful setup (curation and custom instructions), the resulting 15-minute podcasts are often stunningly good, as in mine on SEO:  https://ggl.link/M37yZKk. Here is my https://dub.sh/uo6KzVT explaining a bit more about NoteBook LM. The podcast dazzles, but the more practical use of https://dub.sh/QcXFTmf is as a topic expert one can query. 

Slowly but surely, I’m developing plans to help our region prosper by pooling within an association diverse experts who will work with LLMs and “compeers” to help make our region a model for how to prosper while LLMs take ever more labor time from various professions. Meanwhile, I’d be eager to read here HTers’ best insights and tips. As an example of high-level insight, that above “Advanced SEO” podcast shows, among many other things, a crucial way graphic designers and usability design/writing pros can prosper more now than before LLMs emerged.



Best Regards,



Rob Laporte

CEO  |  R&D Manager

DISC - Making Websites Make Money
mailto:Rob at 2disc.com, 413-584-6500

https://www.2disc.com



NOTE: Emails can be blocked by spam filters throughout the web. If you don’t get a reply within an expected span of time, please call.









_______________________________________________

 Hidden-discuss mailing list - home page: http://www.hidden-tech.net

 mailto:Hidden-discuss at lists.hidden-tech.net

 

 You are receiving this because you are on the Hidden-Tech Discussion list.

 If you would like to change your list preferences, Go to the Members

 page on the Hidden Tech Web site.

 http://www.hidden-tech.net/members
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.hidden-tech.net/pipermail/hidden-discuss/attachments/20250123/52e970b4/attachment-0001.html>


Google

More information about the Hidden-discuss mailing list