ChatGPT - Timely Assistance

 

ChatGPT saved my life, and I’m still freaking out about it
byu/sinebiryan inChatGPT

Keep Thinking!!! 

 

Coding before / After ChatGPT

Coding before ChatGPT:

• Read instructional blogs and official documents.
• Select a relevant helpful tutorial post.
• Set up coding environment needed.
• Execute available blog example code.
• Search StackOverflow for bug solutions.
• Locate and use functioning examples.

Coding after ChatGPT:

• Ask specific coding-related questions.
• Receive comprehensive, detailed explanations.
• Arrange and configure necessary infrastructure.
• Implement and test example code.
• Query errors, and receive alternative solutions.

Observations with ChatGPT:

• Google searches for information are reduced.
• StackOverflow usage noticeably decreased.
• Overall, I experienced time savings.

Keep Exploring!!!

Introducing ChatGPT Enterprise

ChatGPT Enterprise

• You own and control your business data in ChatGPT Enterprise
• We do not train on your business data or conversations, and our models don’t learn from your usage

ChatGPT Enterprise is available today

Advanced Data Analysis (ChatGPT Enterprise version)

Advanced Data Analysis (ADA) has been upgraded to include three new capabilities aimed at enhancing the analysis of text-rich documents: 

• Synthesis -  Analyze information from documents to generate new content or insights
• Transformation - Alter the presentation of information without changing its underlying essence
• Extraction - Identify and pull out specific pieces of information from a document
• Supporting Formats - PDF (.pdf), Text (.txt), PowerPoint (.ppt), Word (.doc), Excel (.xlx), Comma-separated values (.csv)

Contact OpenAI sales team

Advanced-Data Analysis (ChatGPT Enterprise version) - pull out specific pieces of information from a document, Supports - PDF (.pdf), Text (.txt), PowerPoint (.ppt), Word (.doc), Excel (.xlx), Comma-separated values (.csv)

Keep Exploring!!!

ChatGPT

• Decoder-only model.
• Decoder-only architecture does not have an explicit encoder to summarize / context vector the input information
• In Decode only, input sequence is directly fed into the decoder, which generates the output sequence by attending to the input sequence through self-attention mechanisms.

Ref - Link1, Link2

Transformer Key blocks

• In the attention step, words “look around” for other words that have relevant context and share information with one another.
• In the feed-forward step, each word “thinks about” information gathered in previous attention steps and tries to predict the next word.

Ref - Link

What we know about transformers

"What differentiates the Transformer from its predecessors is it’s ability to learn the contextual relationship of values within a sequence through a mechanism called self-attention.

Transformers can be generally categorized into one of three categories:

- encoder onlya la BERT,

- decoder only a la GPT and

- having an encoder-decoder architecture a la T5

Ref - Link

Keep Exploring!!!

Reduce your time to Read / Listen / Summarize / Save time

Reduce your time to Read / Listen / Summarize / Save time

1. Pick the video to summarize - Link

2. Get the transcript from - anthiago

3. Use GPT4 to get key summary

Extract top 10 key predictions from below text. Summarize each line in 10 words only

1.

2.

3.

The Results are

• By end of the next year, chat ChatGPT will be available on mobile phones without internet.
• The figure out mechanism is about making humans scale, akin to an 'analyst with a bad memory'.
• The future of AI is pushing internet intelligence to the edge for every person, country, culture, and company.
• Most popular language models in the world, gptj and neon x, are paving the way for deploying AI systems on-premises.
• Stable diffusion technology can reduce 100,000 gigabytes of images to a 2 gigabyte file that can generate anything.
• Transforming language into images and visuals; four of the top 10 App Store apps based on this premise.
• The second-generation model of this technology will produce photorealistic images and video.
• The trajectory indicates that in five years there will be no need for programmers.
• Open-source models will overtake proprietary models as they offer ownership and interpretation advantages.
• By next year, the capacity of chat GPT will increase, enabling the passage of a level 3 programmer exam

Points I tend to agree with, highlighted in a different color

Earlier I used to spend more time reading complete tech talks, and webinars. Now scanning a few mins and the view count convey intent to read. Passed the video link to generate a transcript. Leverage chatGPT to summarize key points. These five key points and my comments added

• Very soon - Edge-optimized offline LLM models ChatGPT will be available on mobile phones without the Internet. - Edge optimized / Like keyboard word suggestion sentence/reply recommendation possible
• Open-source models will overtake proprietary models as they offer ownership and interpretation advantages. - Now we see Google hosting an unlearning competition for models, More models more explainability/reasoning. The VGG moment is crossed. Resnet / RNN / Another transformer moment coming
• The trajectory indicates that in five years there will be no need for programmers. - This is scary. The need to memorize syntax is no longer needed. How we connect the dots, and consider aspects of the solution becomes key
• Stable diffusion technology can reduce 100,000 gigabytes of images to a 2 gigabyte file that can generate anything. - Right now vision / 3D is very early stages. Next / Coming years 1~2 will get realistic / production version quality with more data trained and larger models tuned to prompts

Keep Exploring!!!

Promises and Lies of ChatGPT - understanding how it works

Key Notes

Basics

• ChatGPT is the idea of n-gram models
• Given n-1 words guess nth word likely to be
• Distribution is learnt from sequence
• People tried in small values of n
• Sample from distribution of words
• More likely words more often

With large data

• Any N, Words next word
• Frequency, Conditional probability
• Generate words if the first word given
• More likely words + Patterns

Large sentences/meanings

• Abstract sequences
• Different answers every time
• Every sequence may be different generated distributions but a similar context is possible
• Chatgpt = something well written

Why it works?

• We believe in what seems realistic
• Connect to human experience
• Fact is different from possibility
• Plausible or probable or reasonable answers 

Similarity to humans

• Humans are not always factual
• It can be perception based
• People can be finalized in civil society
• Machines can suggest without knowing the consequences
• Automation still may have a bias
• Being close to the truth we are impressed

Predictive modeling

Train / predict

Conditional modeling

• Can create bias in information
• Discriminate learning learns a conditional model
• Classifier then finds dogs vs generates dogs both different

Generative distribution - Joint distribution

• The prior distribution of reasonable images
• Teacher = Generative model
• Learning generative model is costlier

The human brain works by on-demand stitching

• chatgpt does something similar
• All learning is compression
• All learning is lossy compression
• jpeg lossy - approximating
• Representation of compressed details
• Significant footprint available to train systems

Good writing for all

• Picaso style pics
• Shakespeare style writing
• Racial profiling not required
• Character and form are not connected
• Generalizations help for survival
• AI as creator / editor

Badly written with original thought is human writing

• Harder to write original creative ways
• Original vs Derivative thinking
• Bad handwriting vs Good content
• Bad package vs Good product
• We have one scale good or bad
• LLM learns from human language
• Most likely completion given soceity is
• Social Enginner on Data

Is this a good representation of all ethnicity ? 

How it for fine tuned ?

• RHLF
• Show results
• asks someone their likes
• Thumbs up / down to change distribution
• Re-learning it
• Collectively offensive content on web vs making a decent prompt engine

• Align to human values
• Concentration campus, Genocide - Human values
• Retrain for cultural norms
• False positive
• Different narrative, different takers

• Make LLM overwrite conditional network through prompts
• Adverserial learning prompts
• How to put knobs how it behaves well

AI systems to work with

• Basically put people to think about problem
• With enough eye balls every downside can be shallow bug
• We need more eyeballs to decide
• ChatGPT will not generate grammatically incorrect sentence
• Core problem of intelligent behavior - planning, diagnosis, reasoning

Keep Exploring!!!

State of GPT

Brilliant talk on LLMs

• Emerging Recipe to train

• Pre-training - compile time of 99%, internet scale dataset
• Data mixture crawl, high-quality data, mixed up, sampled in proportion

• Tokenization
• Text to int representation
• Similar to embedding, word2vec we discuss

• Params
• Token size
• predict the next integer sequence
• 1.4 Trillion Tokens

• hyper parms

• Pretrain - Tokens to Data batches

• Probability distribution of what comes next

• Low loss higher correct probability
• Learn powerful general representations

• LLM + Few short learning is practice
• Transformer forced to multitask in next token
• Forced to understand text, causes
• Better than finetune / prompt them

• Base models are not systems
• It completes what it fills a document

• Not very reliable
• Supervised finetuning
• Small high-quality datasets
• Human contractors
• Prompt-response collection
• Swapping out the training set
• QnA - low quantity - high quality
• People follow the structure and create responses

• Reward read out tokes
• Quality of each completion
• Reformulate loss function with ground truth

• Reinforcement learning with respect to reward model
• Reinforce for higher probabilities
• base model
• SFT model - supervised fine-tuning (SFT)
• RM model - reward model (RM) training
• RL model - reinforcement learning 

• Model Ranking

• Applications

• Template of article

• Time spent on each token
• Token simulators
• Imitate next token
• Fact based knowledge/parameters
• Large working memory
• Transformer direct access to memory
• Chain of thought
• Prompt it will revisit

• Slow and fast reasoning
• Step by step vs one-step process
• Tree search algorithm

• Chain / Agents
• General techniques
• The sequence of thought/observation

• Ask for good performance
• You are an expert on this Topic
• In data distribution of sci-fi

• Tell a prompt not good at
• Use a calculator, Teach LLM to use tools
• Retrieval only vs memory only
• Reteval Augmented models

• Constrained prompting
• Forcing templates
• Output as json

• base model clamped

• Finetuning
• Human contractors

• Recommendations

• Use cases

Keep Exploring!!!

ChatGPT - Product Ideas

 

Keep Exploring!!!

Personal assistant + Health App + Recommendations

 

Ref - Link

Keep Exploring!!!

ChatGPT - LLM - Vector databases - Document Search - Engines

Text is getting revolutionized everywhere. Search, Summarize, Content enrichment, etc..

• Azure Cognitive Search, you can search through millions of documents or data points.
• Azure OpenAI service uses its language modeling capabilities to understand the question and the context provided by the search engine, to generate an answer in natural language.

Ref - Link

Keep Exploring!!!

GenAI based Teaching

Teaching customized with avatars, videos, and images.

Ref - Link

Keep Exploring!!!

Business of ChatGPT

• LLM which can collect more data and the relevant answer will lead the race
• Data is king, relevant information is key
• Creating very special is they key, crawling open web vs limited access to information
• Data race - who owns, who shares, who collaborates

Profits vs Morality. In terms of money vs having a good quality society - Attend important things in life - Link 

Keep Exploring!!!

GPT - Domain - Adaptation - Finance - BloombergGPT

BloombergGPT: A Large Language Model for Finance - BloombergGPT, a 50 billion parameter language model that is trained on a wide range of financial data. This model will help to get information from filings, reports, results with higher relevance because customized to finance domain.

• Social Media NER: The goal of this task is to identify entities that occur in English financially-relevant social media content
• Filings NER: The goal of this task is to identify entities that occur in mandatory financial disclosures filed by companies.

Reads - Link

Demo - Link

Keep Exploring!!!

Think before a #Chat GPT Demo

A tech demo is great but

• Learn how the model works
• Where it may work, where it may fail
• Know its improvements before prior versions
• Have a balanced view
• Doing vs knowing will end up in a great demo but poor domain visibility

Keep Thinking!!!

ChatGPT, LLM, Adoption

Tech Adoption and Other Challenges

From the post, Customizing adding my perspectives 

Interesting Summary, Some key points I liked from the list

• They are useful as writing aids.
• Better systems will come
• Current LLMs should be used as writing aids - Content Writing Systems
• People will use them for what they are helpful with

I hope a few more things can be added to list

• Transparency around which sources of data
• Explainable Answers - transparency around how the model arrived
• Regulating models for fair use, Intentional misuse to change facts/relevance 

Already products are on the market

• Tome - Generative storytelling
• runwayml - GAN Vision / Image / Text Models
• jasper.ai - Get high-quality copy written fast with AI

Future could be Bring your own data - Build your own model. In every domain, Healthcare, Banking, Fintech, History, Geography, and Aerospace we could build custom models which could be used for assistance/learning / generating code/facts/formulas. AI assistants as Teaching assistants :)

"In a few years, people will be comfortable interacting with models and even trust them as their main source of guidance / know by questioning"

Another interesting post on the current shortcomings, I would apply the same to bridge with AI

Bridging the Divided Population - The Big Gap

• We see that of around 6,900 languages currently alive
• Just 291 have Wikipedia
• That is just 4,2% of all languages

The Big Opportunity

• Use Vision / OCR to Digitize all the literature
• Use Neural Machine Translation to translate and increase more Digital Artifacts / Promote more reading / diverse topics
• Create new artifacts, Reach broader population
• Embrace AI tech to aid education in developing countries

In Long Term

• Finetune models to understand social biases such as gender, race, religion

To go far, it's important to go and grow together :)

Most LLM are from the below set of companies

Ref - Link

Keep Exploring!!!

GPT - The Future - New Products vs Threats

Two Posts two interesting views of #ChatGPT #LLM

The future of conversational AI: connecting LLMs to company-specific data - 

Key Summary

• Current focus is researching and developing ways to integrate LLMs like GPT-3 and others into our conversational AI platform. 
• Personalization of LLM for customer data, Game-changer for us is our ability to connect an LLM directly to company-specific data like a website and use it to suggest content
• Increasing transparency around the origin of the generated content

My perspective - Next Phase is Custom models based on domain data - Banking, Finance, Retail, and Healthcare could fuel more personalized systems in indexing / researching domain-relevant data. Companies with data can partner to build models and make money with the models :)

Not all Rainbows and Sunshine: the Darker Side of ChatGPT 

Key Summary

• Transparency around which sources of data
• Explainable Answers - transparency around how the model arrived at this output is lacking
• Potential misuse/manipulation of facts - LLMs could be fine-tuned to spread misinformation and threaten political stability.
• Possibility of plagiarism 
• The enormous carbon footprint of the training lifecycle of an LLM
• Tomorrow a model can be trained to manipulate history / Facts :(

Valid questions, I hope to see more talks around this.

As we keep discussing yes/no there are products already in the market. Using GPT to generate content writing, payment is pay per wordcount :)

Interesting AI Startups

• Tome - Generative storytelling
• runwayml - GAN Vision / Image / Text Models
• jasper.ai - Get high-quality copy written fast with AI

Create hybrid models, Capture the market :), and let the debate continue.

That's a great example of the combination of 3 powerful tactics:

• Freemium model
• Product-led growth
• Community-led growth

Keep Exploring!!!

Learn Istio with ChatGPT

What is istio in layman terms

Istio is a tool that helps you manage and secure the traffic between different services in your application. It allows you to control how traffic flows between services, enforce security policies, and monitor the health of your application. It also helps you to quickly identify and fix any issues that may arise. In layman terms, Istio is like a traffic cop for your application, helping you to keep everything running smoothly.

List basic examples of istio codes with explanation

• DestinationRule: A DestinationRule is a configuration object that defines policies for routing traffic to a service. It can be used to configure advanced features such as traffic splitting, fault injection, and connection pooling.
• VirtualService: A VirtualService is a configuration object that defines how traffic should be routed within an Istio service mesh. It can be used to route requests to different versions of a service, or to route requests to different services based on the request headers.
• Gateway: A Gateway is a configuration object that defines a load balancer for incoming requests. It can be used to configure external access to the services in an Istio service mesh.
• ServiceEntry: A ServiceEntry is a configuration object that defines an external service that is accessible from within the Istio service mesh. It can be used to define external services that are not part of the Istio service mesh.
• Sidecar: A Sidecar is a configuration object that defines how a service should be deployed within an Istio service mesh. It can be used to configure features such as mutual TLS authentication, traffic mirroring, and request tracing.

How to get hands on with istio

• Install Istio on your local machine: Install Istio on your local machine by following the instructions on the Istio website.
• Deploy a sample application: Deploy a sample application such as the Bookinfo application to get hands-on with Istio.
• Configure Istio: Configure Istio for your application by setting up routing rules, policies, and other features.
• Test your application: Test your application by sending requests to it and verifying the results.
• Monitor your application: Monitor your application using Istio’s built-in monitoring tools.
• Troubleshoot any issues: Troubleshoot any issues that arise while using Istio.

Keep Exploring!!!

ChatGPT to meet our Leaders

Write a disagreement dialogue between Tyler Dryden and mahatma Gandhi about inequality

Tyler: "Inequality is a natural part of life and should be accepted."

Mahatma Gandhi: "I disagree. Inequality is not natural, it is a result of the systems and structures we have created. We should strive to create a society where everyone is treated equally."

Tyler: "But that's impossible. People are different and have different needs, so it's impossible to treat everyone equally."

Mahatma Gandhi: "No, it's not impossible. We can strive to create a society where everyone is given the same opportunities and access to resources, regardless of their background or identity. That is the only way to truly achieve equality."

Write a disagreement dialogue between karlmarx and socrates on equality

Karl Marx: Equality is an illusion. The working class will always be oppressed by the ruling class, no matter how much we strive for equality.

Socrates: I disagree. Equality is achievable if we all work together to create a more equitable society. We can create a world where everyone is treated equally, regardless of their social class.

Karl Marx: That may be true in theory, but in practice, it's impossible. The ruling class will always find ways to maintain their power and privilege.

Socrates: I don't think it's impossible. We can create a society where everyone has an equal opportunity to succeed. We just need to create policies that promote fairness and justice.

Write a summary of why lack of education is dangerous for future

Lack of education is a dangerous issue for the future of our society. Without education, individuals are unable to access the knowledge and skills needed to succeed in the modern world. This can lead to poverty, unemployment, and social exclusion. Furthermore, without education, individuals are unable to make informed decisions about their lives, leading to a lack of civic engagement and a lack of understanding of the world around them. This can lead to a lack of economic opportunity, a lack of political engagement, and a lack of understanding of global issues. Without education, individuals are unable to contribute to the advancement of society, and this can have a detrimental effect on our collective future.

Write a dialogue between karl marx and socrates in 1000 words

Karl Marx: Greetings, Socrates! It is an honor to meet you.

Socrates: The honor is mine, Karl. What brings you here?

Karl Marx: I have come to discuss the nature of justice.

Socrates: Ah, justice. A topic that has been debated by philosophers for centuries. What is your opinion on the matter?

Karl Marx: I believe that justice is a product of the economic and social conditions of a society. In other words, justice is determined by the economic and social structures of a society.

Socrates: Interesting. Can you explain further?

Karl Marx: Certainly. I believe that the economic and social conditions of a society determine the level of justice that is available to its citizens. For example, in a capitalist society, the wealthy have more access to justice than the poor. This is because the wealthy have more resources and power to influence the legal system.

Socrates: That is an interesting perspective. What do you think can be done to ensure that justice is available to all citizens, regardless of their economic and social status?

Karl Marx: I believe that the only way to ensure justice for all is to create a society that is based on equality and fairness. This means that all citizens should have equal access to resources and opportunities. Furthermore, the legal system should be structured in such a way that it is fair and impartial.

Socrates: That is a noble goal. Do you think it is possible to achieve?

Karl Marx: I believe that it is possible, but it will require a great deal of effort and dedication. We must work together to create a society that is based on justice and equality.

Socrates: I agree. We must strive to create a society that is just and equitable for all citizens.

Karl Marx: Indeed. Thank you for your time, Socrates. It has been a pleasure discussing justice with you.

Socrates: The pleasure is mine, Karl. I wish you the best of luck in your endeavors.

Keep Exploring!!!

Five things About GPT

Five Things about GPT

#1. GPT-3, or the third generation Generative Pre-trained Transformer, is a neural network machine learning model trained using internet data to generate any type of text

ChatGPT is a chatbot technology developed by 𝐎𝐩𝐞𝐧𝐀𝐈. It is designed to assist with a variety of tasks and functions, including answering questions, providing information, and completing tasks. 

#2. Number of layers and parameters

#3. Parameters

• GPT-2 was released in February 2019 with 1.5 billion parameters 
• GPT-3 was released in June 2020 with 175 billion parameters (~120x improvement)
• GPT-4 will be released soon and is expected to have 100 trillion parameters (~500x improvement)

#4. How intelligent can it be?

#5. GPT-4 is built on the Transformer architecture, which has been effective for a variety of machine-learning tasks, including computer vision. This means that GPT-4 might be used for tasks such as image and video generation

#6. Training Approach

Keep Exploring!!!