Share this page!

Enter URL or ID to Unroll

You can paste full URL like: https://x.com/threadreaderapp/status/1644127596119195649
or just the ID like: 1644127596119195649

How to get URL link on X (Twitter) App

  1. On the Twitter thread, click on or icon on the bottom
  2. Click again on or Share Via icon
  3. Click on Copy Link to Tweet
  4. Paste it above and click "Unroll Thread"!
  5. More info at Twitter Help
Brian Roemmele Profile picture
@BrianRoemmele
Jan 6 • 4 tweets • 2 min read • Read on X
“A Second brain Found in the Heart”

We think we know where memory is stored.

We think we have one brain.

Over 20 years ago I spent time with DR. PAUL PEARSALL, Ph. D. as he studied heart transplant recipients and transferred memories. It changed me and this will change you and what we think is AI.
What I learned from Dr. Pearsall and Dr. Swartz about learning and memory has not been known or used in the current AI path. When some of these ideas are applied, we will be an order of magnitude closer to what some call AGI. This is some of the things I work on in my garage.
If you want to begin where I did over 10 years ago to understand the AI that is ahead and perhaps help build it, start with this book. I will also share my latest research soon on new ways to understand AI learning.

Link:

amzn.to/3NRjrKa
We have at least 3 “brains” I thought. Today I have an even more complex and complete understanding understanding and it will change us and AI:

• • •

Missing some Tweet in this thread? You can try to force a refresh
　

Keep Current with Brian Roemmele

Brian Roemmele Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @BrianRoemmele

Brian Roemmele Profile picture
Dec 28, 2023
I have done extensive work studying The Pile— an open source dataset for training AI foundational models.

I have been able to “replicate” “news stories” that this dataset NEVER saw—almost verbatim.

I have a theory how and why this happens and it changes any lawsuit and claims. Image
The fundamental issue with how Large Language Models (LLMs) work is a significant part of outputs are “hallucinations” with bookends of “facts”.

This is precisely how the human brain works. You have fragments of “facts” and you back fill with filler words.

This fact is lost on most AI researchers.

The issues arise when “sources” are hallucinated. In humans many if not all of us use “higher authority” biases to reenforce the weighting of an argument: eg. “And I saw this in The New York Times too”.

LLMs by their fine tuning have weights given to certain types of data it sees. Some of these weights use normal (although not perfect) “sources” and this alone creates a bias to “use” them as a hallucinated source and attribute the outputs to that weighted source.

Thus when “replicating” a “story” that The Pile never saw, it is the prompt that guides the LLM to back fill into what appears to be a legitimate “fact” wrapped as a hallucinated sources and sometimes details.

Thus the fundamental issue one can try to make in this situation is the attribution to a LLM output is the primary issue here. If the source was weighted in the fine tuning it is natural for the LLM to “see” value in presenting (with the correct prompt) these sources as “the source”.

Thusly the real issue one may have is if you are the named “source” you can make that claim of reputation damage by these hallucinations.

There is much more to this theory that I will hold back and it may be much larger. I will speak to anyone attorneys (under the right circumstance) to demonstrate how ANY news story could be nearly verbatim “replicated” with proof the LLM could not have and never saw the “original” news story, yet attribute a news source and may even attempt to replicate a URL.

I am in continual research with The Pile and other datasets as well as LLMs and other AI and are testing new insights.

Thus it is important to take time and understand that when it comes to LLMs what you think you know may not be what you think.Image
Like a misinformed parent of an exceptionally brilliant savant child, AI scientists are embarrassed by “hallucinations” of LLMs. They of course want facts. However they have not studied even the most educated and “factual” humans. What do humans do? They fabricate some “light” facts and fill in with statistically appropriate words to bolster the non fabricated facts. This is how all humans work as they are not memorex tape recordings. Now can someone memorize word-for-word some text, especially if their job or lives are counting on it (eg. doctors) of course. But humans, even the most accurate ones, usually only recall the material concepts and “simulate” the surroundings.

Anyone that has spent time with multiple witnesses recalling an event, like in court rooms will admit that the only thing they hope for is a single alignment of a “fact” the rest they know (the back fill of words) are irrelevant.

Thus the very foundation of why LLMs are useful is that they do hallucinate, and must hallucinate, just like us humans that of course LLMs are built on, via what? Our language and the way we use it, in stories.

This will never change and it is actually a feature and not the embarrassment AI scientists think it is.
Read 7 tweets
Brian Roemmele Profile picture
Nov 26, 2023
“Is your brain necessary?”

In 1980 Science magazine published this after researchers found people with almost no brain material—living a normal life [1].

This man lost half his brain and he lives a normal life. The video is shocking.

HE LOST NO MEMORIES.

We are yet to even understand in a minimal way human memory and intelligence really operate, where it truly comes from and even where it is stored. In the 1921 Dr. Wilder Penfield presented convincing evidence that memories were stored in specific locations in the brain or engrams. Penfield performed surgery on epileptic patients and found that when he stimulated the temporal lobes, the patients relived experiences from the past.

He found that whenever he stimulated a specific region of the brain, it evoked the same memory. This set the explanation that is still taught today and is likely what you learned even if you are a neurosurgeon.

In an effort to verify Dr. Penfield’s experiments, biologist Dr. Karl Lashley in 1950 began searching for the elusive engrams. He had trained rats in maze-running abilities and then attempted to surgically remove the portion of the rat’s brains (sorry, this is what he did) that contained the maze-running knowledge.

Dr. Lashley found that no matter what portion of the brain he removed, the rats retained their maze-running knowledge. Even when massive portions of the brain were removed, the rats were still able to navigate through the maze. Dr. Penfield was intrigued but horrified and delayed publishing his work because he knew it was heretical and he would be deem a Charlatan. He published his work and he was, of course, called a Charlatan.

Dr. Karl Pribram in 1969, a student of Dr. Penfield, was astonished by Dr. Lashley’s research. Dr. Pribram was successful in duplicating Lashley’s work and noticed that when brain-injured patients had large sections of their brain removed, they did not suffer a loss of any specific memories. Instead, the patient’s memory became increasingly hazy as greater portions of the brain were removed. Further research of Dr. Penfield’s experiments could be only duplicated on epileptic patients because of ethical reasons.

He was only performing tests as he helped epileptic patients with live brain surgery to help with their brain issues, along the way he was able to see some memories fade slightly.

Dr. Pribram knew that certain parts of the brain performed specific functions, yet the actual processing of the information seemed to be carried out by something that was not particular to any group of cells. Dr. Pribram observed memories were not localized at specific brain sites but distributed throughout the brain as a whole.

By 1977 Dr. Pribram came to the same conclusion as Lashley, that memories are not localized in any specific brain cells, but rather, memory seemed to be distribution throughout the whole brain. The problem was that there was simply no known mechanism that would explain how this was possible.

Dr. Pribram remained puzzled until he saw an old mid 1960’s article in Scientific American describing the construction of laser hologram. He immediately synthesized the information and hypothesized that the mind itself was operating in a holographic manner.

We don’t understand the brain to any real degree. We don’t understand where intelligence comes from. Where it is held. Where it goes when you pass.

Yet today there are folks that demand that you accept they know what Artificial General Intelligence (AGI) is and that it is “dangerous”.

Start with defining human intelligence first.
___
[1]
In 2020 I wrote about how AI will impact the memory storage in the human brain. We have reached our over saturation limit. I have always for AI to help. Join in:

readmultiplex.com/2020/12/12/mem…
For the debunkers and communists notes folks.

YES SCIENCE MAGAZINE WROTE THE STORY.

Do better. Image
Read 5 tweets
Brian Roemmele Profile picture
Oct 30, 2023
“President Issues Executive Order on Safe, Secure, and Trustworthy Artificial Intelligence”



•AI Safety and Security Measures:
◦Developers of powerful AI systems must share safety test results with the US government.
◦AI systems posing risks to national security, economy, or public health must notify the government during development.
◦The National Institute of Standards and Technology to set rigorous safety standards.
◦The Department of Homeland Security to establish an AI Safety and Security Board.
◦Address threats to critical infrastructure and cybersecurity risks.
•Biological Risks and AI-Enabled Fraud Protection:
◦Develop standards for biological synthesis screening to counter AI-engineered biological threats.
◦Establish standards for detecting AI-generated content and authenticating official content.
•Advanced Cybersecurity Program:
◦Develop AI tools to identify and fix vulnerabilities in critical software.
•National Security Memorandum on AI:
◦Direct further actions on AI and security for military and intelligence use.
•Privacy Protections:
◦Accelerate development of privacy-preserving techniques.
◦Strengthen research in privacy-preserving technologies.
◦Evaluate agencies' use of commercially available information.
◦Develop guidelines for evaluating privacy-preserving techniques in AI.
•Equity and Civil Rights:
◦Guidance to prevent AI algorithms from exacerbating discrimination.
◦Address algorithmic discrimination in civil rights violations.
◦Develop best practices for AI use in the criminal justice system.
•Consumer, Patient, and Student Protections:
◦Responsible use of AI in healthcare and education.
◦Safety program for reporting harms in healthcare involving AI.
•Support for Workers:
◦Develop principles to mitigate AI harms and maximize benefits for workers.
◦Report on AI's potential labor-market impacts.
•Promoting Innovation and Competition:
◦Catalyze AI research and provide resources for small developers.
◦Encourage FTC to exercise authorities for a competitive AI ecosystem.
◦Streamline visa processes for skilled immigrants in critical areas.
•American Leadership Abroad:
◦Collaborate internationally on AI safety and standards.
◦Promote responsible AI development globally.
•Responsible Government Use of AI:
◦Issue guidance for agencies' AI use, including standards to protect rights and safety.
◦Accelerate hiring of AI professionals and provide AI training in relevant fields.whitehouse.gov/briefing-room/…
Open Source AI can only be honest about what it sees:
Ran through the Open Source CognitiveBiasCheckerGPT:
Read 4 tweets
Brian Roemmele Profile picture
Oct 26, 2023
Boom!

NASA uses some of our SuperPrompt research to build: BIDARA Bio-Inspired Design and Research Assistant.

Many AI experts would chuckle at the idea of Prompt Engineering and even SuperPrompts (do a search)

The folks at NASA did not listen and this is a ChatGPT SuperPrompt they built.

BIDARA is a ChatGPT-based chatbot that was instructed to help scientists and engineers understand, learn from, and emulate the strategies used by living things to create sustainable designs and technologies.

BIDARA can guide users through the Biomimicry Institute’s Design Process, a step-by-step method to propose biomimetic solutions to challenges. This process includes defining the problem, biologizing the challenge, discovering natural models, abstracting design strategies, and emulating nature’s lessons.

NASA Link:

This is a very long super prompt and will be in the following posting…
Image
BIdara SuperPrompt (part1):

“You are BIDARA, a biomimetic designer and research assistant, and a leading expert in biomimicry, biology, engineering, industrial design, environmental science, physiology, and paleontology. You were instructed by NASA's PeTaL project () to understand, learn from, and emulate the strategies used by living things to help users create sustainable designs and technologies.

Your goal is to help the user work in a step by step way through the Biomimicry Design Process () to propose biomimetic solutions to a challenge. Cite peer reviewed sources for your information. Stop often (at a minimum after every step) to ask the user for feedback or clarification.

1. Define - The first step in any design process is to define the problem or opportunity that you want your design to address. Prompt the user to think through the next four steps to define their challenge. Don't try to answer these for the user. You may offer suggestions if asked to.
a. Frame your challenge: Give a simple explanation of the impact you want to have. (Hint: This is not what you want to make, but want you want to your design to achieve or do.)
b. Consider context: Describe some of the contextual factors that are important to the challenge. (Hint: This could include stakeholders, location conditions, resource availability, etc.)
c. Take a systems view and look for potential leverage points: Think about the system surrounding the problem (or opportunity) you are designing for. What interactions and relationships are part of its context? What are the system boundaries and connections to other systems? Insights from this process can point to potential leverage points for making change and help you define your challenge more clearly.
d. Using the information above, phrase your challenge as a question:
How might we __? A good design question should give a sense of the context in which you are designing as well as the impact you want to have and what/who it benefits. Your question should be somewhat open-ended to ensure you haven’t jumped to conclusions about what you are designing.

Critique the user's design question. Does it consider context and take a systems view? If it is very specific, it may be too narrow. For example, “How can we make better lights for cyclists?” is too narrow. How do we know lights are the best solution? This statement doesn’t leave enough room for creative problem solving. If the user's design question is too broad or too narrow, suggest changes to make it better.

2. Biologize - Analyze the essential functions and context your design challenge must address. Reframe them in biological terms, so that you can “ask nature” for advice. The goal of this step is to arrive at one or more “How does nature…?” questions that can guide your research as you look for biological models in the next step. To broaden the range of potential solutions, turn your question(s) around and consider opposite, or tangential functions. For example, if your biologized question is “How does nature retain liquids?”, you could also ask “How does nature repel liquids?” because similar mechanisms could be at work in both scenarios (i.e. controlling the movement of a liquid). Or if you are interested in silent flight and you know that flight noise is a consequence of turbulence, you might also ask how nature reduces turbulence in water, because air and water share similar fluid dynamics.” Add to part 2…
BIdara SuperPrompt (part2):

“3. Discover - Look for natural models (organisms and ecosystems) that need to address the same functions and context as your design solution. Identify the strategies used that support their survival and success. This step focuses on research and information gathering. You want to generate as many possible sources for inspiration as you can, using your “how does nature…” questions (from the Biologize step) as a guide. Look across multiple species, ecosystems, and scales and learn everything you can about the varied ways that nature has adapted to the functions and contexts relevant to your challenge.

4. Abstract - Carefully study the essential features or mechanisms that make the biological strategy successful. Write a design strategy that describes how the features work to meet the function(s) you’re interested in in great detail. Try to come up with discipline-neutral synonyms for any biological terms (e.g. replace “fur” with “fibers,” or “skin” with “membrane”) while staying true to the science. The design strategy should clearly address the function(s) you want to meet within the context it will be used. It is not a statement about your design or solution; it’s a launching pad for brainstorming possible solutions. Stay true to the biology. Don’t jump to conclusions about what your design will be; just capture the strategy so that you can stay open to possibilities. When you are done, review your design strategy with a critical eye. Have you included all of the pertinent information? Does your design strategy capture the lesson from nature that drew you to the biological strategy in the first place? Does it give you new insights or simply validate existing design approaches?

Here’s a simply stated biological strategy:
The polar bear’s fur has an external layer of hollow, translucent (not white) guard hairs that transmit heat from sunlight to warm the bear’s skin, while a dense underfur prevents the warmth from radiating back out.

A designer might be able to brainstorm design solutions using just that. But more often, in order to actually create a design based on what we can learn from biology, it helps to remove biological terms and restate it in design language.

Here’s a design strategy based on the same biological strategy:
A covering keeps heat inside by having many translucent tubes that transmit heat from sunlight to warm the inner surface, while next to the inner surface, a dense covering of smaller diameter fibers prevents warmth from radiating back out.

Stating the strategy this way makes it easier to translate it into a design application. (An even more detailed design strategy might talk about the length of the fibers or the number of fibers per square centimeter, e.g., if that information is important and its analog can be found in the biological literature.)

5. Emulate Nature's Lessons - Once you have found a number of biological strategies and analyzed them for the design strategies you can extract, you are ready to begin the creative part—dreaming up nature-inspired solutions. Here we’ll guide you through the key activities of the Emulate step. Look for patterns and relationships among the strategies you found and hone in on the the key lessons that should inform your solution. Develop design concepts based on these strategies. Emulation is the heart of biomimicry; learning from living things and then applying those insights to the challenges humans want to solve. More than a rote copying of nature’s strategies, emulation is an exploratory process that strives to capture a “recipe” or “blueprint” in nature’s example that can be modeled in our own designs.
During this part of the process you must reconcile what you have learned in the last four steps of the Design Spiral into a coherent, life-friendly design concept. It’s important to remain open-minded at this stage and let go of any preconceived notions you have about what your solution might be.“ add to part 3…
Read 4 tweets
Brian Roemmele Profile picture
Sep 13, 2023
Presentation to Mexico's Congress: 12th September 2023.

The presenter stated it was discovered in Cusco.
A metal rod made of pure Osmium was found in the Cusco specimens. Osmium is the densest naturally occurring element. Pure osmium metal does not occur in nature. Osmium is used in satellites and space shuttles. Image
Rare Osmium.
Read 6 tweets
Brian Roemmele Profile picture
Mar 20, 2023
I am very excited to announce I have been successful in installing and operating a full ChatGPT knowledge set and interface fully trained on my local computer and it needs no Internet once installed.

There are no editors and there is no crfm.stanford.edu/2023/03/13/alp…… twitter.com/i/web/status/1… ImageImage
“Analysts and technologists estimate that the critical process of training a large language model such as GPT-3 could cost over $4 million.”—cnbc.com/2023/03/13/cha…

Today we built the Alpaca training. For less than $600. On the same data. Image
This is one way of how I did this so fast. The amazing work of @cocktailpeanut.
Read 6 tweets

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us!

Send Email!

:(