LLM, AI and ML

[TargeT]

Posted by Vangelo (here)
Hi TargetT, I am very much looking forward to the results with your mom. When you say that you will call her, are you having a phone call or a texting session? If a phone call, she will know its not your voice.

I might even be able to do it with video (but probably not with my mom), this is pretty insane.

Source: https://youtube.com/watch?v=p_xkBMnQVI0

This guy is an AI youtuber and he was tricked by video... haha

[TargeT]

I've talked to a couple lawyers in person and those kind enough to spare some time for my situation re: legal advice on a medical discrimination lawsuit against amazon.

I've talked with 3 firms that specialize in that (though only 2 lawyers and one nice young lady that wanted 500 for a consultation with the lawyer) and have searched with what little knowledge I have. I felt something was off, though not early enough; but I did collect a bit of emails and screenshots from slack.

I'm going to allow AI to do everything from now on (at least, until it requires a lawyer); so far it seems like the same thing.

[ExomatrixTV]

Great Thread ...

For all new people, you may also want to study the latest 20+ posts here

cheers,
John 🦜🦋🌳

[TargeT]

Posted by TargeT (here)
I'm going to allow AI to do everything from now on (at least, until it requires a lawyer); so far it seems like the same thing.

I uploaded a few of the emails and screen shots and asked it for a timeline

Certainly. Here is the revised timeline with source references for a potential medical discrimination lawsuit:

### Comprehensive Timeline for Medical Discrimination Lawsuit with Sources

**Early June 2023**:
- T experiences a toe injury (misdiagnosed), noted as the third occurrence in June.
- Source: [June toe, third occurrence in June.pdf]

**July 6, 2023**:
- Supervisor boss sends T information on leave options and accommodations in response to Erik's health concerns.
- Source: [EmailFromBoss_FMLA.txt]

**Early July 2023**:
- T starts to face severe medical issues, suspected to have begun before June.
- Source: [reported medical issues in early July.docx]

**July 19, 2023**:
- T is admitted to the ER for emergency surgery due to arterial thrombosis.
- Source: [MedicalApointmentCommunication.JPG]

**July 29, 2023**:
- Released from the ICU after surgery for Arterial Thrombosis; commences treatment involving multiple specialists.
- Source: [MedcomsLOAconfusion.JPG]

**August 17, 2023**:
-T discusses taking a personal day due to complications with blood thinner medication.
- Source: [Aug, blood thinner issue.JPG]

**August 18, 2023**:
- Meeting with supervisor boss is rescheduled due to T medical needs.
- Source: [Rescheduled meeting due to medical issues.png]

**August 20, 2023**:
- T receives an email from Boss with resources for leave and accommodations, sent hours after surgery.
- Source: [EmailFromBoss_FMLA.txt]

**August 22, 2023**:
- Receives Pivot Plan documentation outlining performance expectations and deadlines.
- Source: [Pivot Plan document.pdf]

**August 25, 2023**:
- T voices concerns about meeting Pivot Plan expectations amidst ongoing medical treatments.
- Source: [Pivot Plan Concerns Email.pdf]

**August 28, 2023**:
- T sends an email to his legal counsel indicating the medical issues started before June and that his vascular surgeon would attest to that.
- Source: [Legal counsel communication.pdf]

**August 29, 2023** (Proposed Start Date for FMLA):
- Proposed start date for FMLA leave; actual approval is delayed.
- Source: [FMLA Proposed Start Date Email.pdf]

**September 1, 2023**:
- First Pivot Plan deadline; Ts performance potentially impacted by health situation.
- Source: [Pivot Plan First Deadline Email.pdf]

**September 5, 2023**:
- Second Pivot Plan deadline to present a risk document for MW2.
- Source: [MW2 Risk Document Deadline Email.pdf]

**September 27, 2023**:
- Erik expresses confusion about the FMLA leave process and the lack of information provided by HR.
- Source: [FMLA Process Confusion.png]

**September 28, 2023**:
- FMLA leave is officially approved; coincides with the final Pivot Plan deadline.
- Source: [FMLA Approval Email.pdf]

**November 3, 2023**:
- Receives instructions for the appeal process from hr_person.
- Source: [Appeal Process Instructions Email.pdf]

**November 9, 2023**:
- Submits the Appeal Form and Appendix.
- Source: [Appeal Form Submission Confirmation Email.pdf]

**November 10, 2023**:
- Deadline for submitting appeal documentation.
- Source: [Appeal Documentation Deadline Reminder Email.pdf]

**November 14, 2023**:
- Informed of an unsuccessful PiP appeal, with substantial redaction in the appeal documents.
- Source: [Unsuccessful PiP Appeal Notification Email.pdf]

**Notes**:
- The timeline details Ts ongoing health issues and their intersection with critical work performance evaluation periods.
- Communications from atto his employer indicate a clear attempt to inform them of his medical limitations and seek reasonable accommodations.
- The delay in FMLA approval and the lack of clarity provided by HR may point to potential mishandling of Ts situation.
- The substantial redaction of appeal documents raises concerns about the fairness and transparency of the appeal process.

Legal counsel should evaluate this timeline, alongside the referenced sources, to assess the strength of a medical discrimination claim.

Next I'll upload my medical appointments and procedures for it to integrate into the time line, then have it write a letter to EEO for me as well as a letter to amazon renegotiating my severance in light of this situation


Time.spent so far, about 2 hours, mostly sifting through files before I just gave them all to my bot for parsing

[ExomatrixTV]

• What comes after LLMs?

Source: https://youtube.com/watch?v=Hz8yZUXsYrs

00:00 - The Future of A.I.
04:24 - A.I. Stocks to watch

• youtube.com/@AnastasiInTech/videos

[TargeT]

This is not verified information, but the whole shake up at OpenAI might have been due to a huge break through...

Source: https://youtube.com/watch?v=3d0kk88IE8c

and it's seems more and more likely....

Source: https://youtube.com/watch?v=R0PEPzcvWRQ

This is plausible... I knew it would happen too quickly, but this seems... well ... too quickly.

AGI is the current goal and Q* might be the corner stone.

[TargeT]

I'm going to play with this next, I already have kali Linux and VMware player .. thus might be very useful if you like playing with software

Source: https://youtube.com/watch?v=7uRz_bb95YM

[TargeT]

Looks like cryptologists need to get real innovative REAL soon.. it looks like Q* may have figured out on its own how to break complex encryption.

Definitely happening quicker than I thought it would (Not proven, just leaks and speculation so far... but it seems to be very possible). This is all due to the "Deep Reinforcement Learning" where one AI will "grade" the other (until very recently it was all done by humans) so it can self improve beyond human capability.

Source: https://youtube.com/watch?v=2BT9H3YqZqM

It broke AES-192 almost instantly using a method that is not understood yet (if the leak is true). This seems a bit fantastical to me, but at the same time... I think its very possible, especially based on the March findings.

[TargeT]

I was making a joke with my girl.. I used 7 words to get this "photographic evidence" of the situation....

also: this is a throwaway twitter account so I could get access to Xai (Grok), nothing other than a photo hosting platform for me now since free ones are hard to find (I noticed Avalon (smartly) discontinued their hosting); however I gladly give money to the platform as I see it as the best "actual" action that is helpful for free speech.

[Johnnycomelately]

Not sure if this belongs here, but thought it might be of interest to OP TargeT.

I know very little about AI (don’t understand much of the posts above), but found it bemusing to learn of frailties in that discipline.

Nepotistically Trained Generative-AI Models Collapse
Matya ́ˇs Boha ́ˇcek Stanford University maty@stanford.edu
Hany Farid
University of California, Berkeley hfarid@berkeley.edu


Abstract

Trained on massive amounts of human-generated content, AI-generated im- age synthesis is capable of reproducing semantically coherent images that match the visual appearance of its training data. We show that when retrained on even small amounts of their own creation, these generative-AI models produce highly distorted images. We also show that this distortion extends beyond the text prompts used in retraining, and that once poisoned, the models struggle to fully heal even after retraining on only real images.

1 Introduction[

From text to audio and image, today’s generative-AI systems are trained on large quantities of human-generated content. Most of this content is obtained by scraping a variety of online sources. As generative AI becomes more common, it is reasonable to expect that future data scraping will invariably catch generative AI’s own cre- ations. We ask what happens when these generative systems are trained on varying combinations of human-generated and AI-generated content.
Although it is early in the evolution of generative AI, there is already some evi- dence that retraining a generative AI model on its own creation – what we call model poisoning – leads to a range of artifacts in the output of the newly trained model. It has been shown, for example, that when retrained on their own output, large language models (LLMs) contain irreversible defects that cause the model to produce gibberish – so-called model collapse [22].
Similarly, on the image generation side, it has been shown [1] that when retrained on its own creations, StyleGAN2 [12] generates images (of faces or digits) with visual and structural defects. Interestingly, the authors found that there was a deleterious effect on output as the ratio of AI-generated content used to retrain the model ranged from 0.3% to 100%.
It has also been shown that in addition to GAN-based image generation, diffusion- based text-to-image models are also vulnerable. The authors in [15, 14], for example, showed that in a simplified setting, retraining on one’s own creation can lead to…



Here is the 11 page PDF. The above paste is the first page entire. In the last 2 pages, some of the referenced papers have interesting titles.

[I know how to embed PDFs, chose not to, because serious readers will want to bookmark it. And, I’ve found embedded PDFs to be problematic to access]

https://arxiv.org/pdf/2311.12202.pdf

[TargeT]

That article was absolutely correct.
I posted your message to one of my "AI" bots:

Also, I should be clear that I think there is no such thing (currently) as "AI" there are very clever algorithms that seem like it, but are just very very good mathematical tricks.

This is not to say that they are not hugely impactful to our every day lives (this is the metric that interests me... what does it mean to "Daily life")

This is also going to lead to AGI (artificial general intelligence) which can be worrisome due to humanities proclivity of leaning on advancements entirely with out understanding of even the basic aspects of it (who here can explain a 4 stroke engine and it's relationship to timing and spark plug firing cycles to me(the question gave away a lot on it's own)?) which generally leads to at the very least misuse of the tool, the worst... well that's in a lot of fiction.

Until early this year (this thread talks about it) these models (the "AI") were corrected by humans, this is called DRL (Deep Reinforcement Learning). Now we are using different AI models (originally still trained by an honestly terrifyingly small subset of humans) to grade them selves, and a few posts up... there may be (still unverified) situations where AI was solving math problems that were previously considered unsolvable and via methods we do not currently understand; I suppose that doesn't seem like much to most people but if you look at our world economies and how much they re lie on "trust" and encryption this is a potentially horrifying event, maybe enough to get the CEO of OpenAI (chatGPT) fired.....

[TargeT]

Haven't started my Swarm project yet, but here's my general approach (of course, out put by GPT4):

To create a locally hosted AI or LLM (Large Language Model) swarm for personal research that includes local files, you can consider the following methods, based on the latest information available:

1. **GPT4All**: Offers a desktop client for various operating systems, allowing the use of models such as Llama-2-7B chat and OpenAI's GPT-3.5 and GPT-4. It includes a LocalDocs plugin to chat with your own documents locally【20†source】.

2. **LLM Command Line Tool**: Available for Python users, this tool allows the use of open source LLMs locally on your machine without needing to write Python code. It supports models like GPT4All and others through plugins【21†source】.

3. **Ollama for Mac and Linux**: A simple command-line tool to download and run models, including several Llama-based versions, with plans to support Windows【22†source】.

4. **h2oGPT**: A chat desktop application by H2O.ai, offering limited document Q/A capability with Llama models. It's suitable for machine learning novices and can interact with your local files in various formats【23†source】.

5. **PrivateGPT**: Designed to query your own documents in various formats using natural language, ensuring 100% privacy as no data leaves your execution environment【24†source】.

For the most up-to-date methods and tutorials, you may also explore YouTube videos and tutorials on these tools and related topics.

[TargeT]

Googles GEMINI is beating GPT4... This is a good one to watch.

Source: https://youtube.com/watch?v=-a6E-r8W2Bs

Source: https://youtube.com/watch?v=KQBA62yZURc

I'll play around with it and update.

[TargeT]

Hmmm, actual competition.... is this capitalism?

Source: https://youtube.com/watch?v=n6Fj27R7OmQ

Currently the only way to use GEMINI is via https://bard.google.com/

Though I expect this to change soon.

[TargeT]

Got a 3rd job offer based on AI suggestion and resume re-writing... in total I've jumped 30k annually in base salary (from the first offer to this last one) and secured a 22k sign on bonus for this recent one...

This is a very impressive tool.


From all my testing and prompting GPT4 is still vastly superior despite claims otherwise (if you can get signed up, it's vastly worth the 20$ a month if your an information junkie like me).

[TargeT]

Welcome to living in a period of known exponential growth (This has never (that we know of) happened before en mass).

Source: https://youtube.com/watch?v=OSV7cxma6_s

[TargeT]

From all my testing and prompting GPT4 is still vastly superior despite claims otherwise (if you can get signed up, it's vastly worth the 20$ a month if your an information junkie like me).

It appears there is a reason why got is better

Source: https://youtube.com/watch?v=eIIIVHWpMUA

[TargeT]

Posted by Johnnycomelately (here)
Not sure if this belongs here, but thought it might be of interest to OP TargeT.

I know very little about AI (don’t understand much of the posts above), but found it bemusing to learn of frailties in that discipline.

Haha, wait till you learn about AI hallucinations; it is definitely not a simple tool.

Source: https://youtube.com/watch?v=cfqtFvWOfg0

Here is an example of me prompting GPT4 and it clearly showing bias to support Intellectual Property and would not give me a return that I was asking for:

https://chat.openai.com/share/fd3690...f-ffed3a2098df

[TargeT]

Yup, it's verified...

Gpt 4.5 turbo is out and amazing

Source: https://youtube.com/watch?v=rpxIUpwz9U0

[TargeT]

I think this hints that the previous link about *Q was correct... I can see why they are being very cautious

Source: https://youtube.com/watch?v=PxrYD9x3V4E

if encryption can be easily broken our electronic (mostly financial) communications based on trust will be destroyed.

but that's a bit "IF"