If you have been following the news, there is a lot of noise on the rapid improvement of generative AI models such as ChatGPT.
Auto GPT? What if we could have an AI agent that, when given instruction, could operate independently, use APIs like Google Search API, and have access to Make /Zapier to actually "do" software-related activities?
What if you could string together Generative AI models to feed off each other and generate prompts suitable to the activity? You could get your coffee as your AI agent executes the task with minimum supervision.
That is the promise of Auto GPT. A couple of experimental projects are already out on GitHub. Search for BabyAGI, AgentGPT, and AutoGPT by Significant Gravitas on Twitter, and you will get an idea of the immense interest that it has generated. But first, what is Generative AI?
What is Generative AI?
Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and Generative Flow Models. Generative AI is a type of artificial intelligence that uses deep learning techniques such as neural networks to learn patterns and correlations from a dataset and generate new outputs. Popular generative AI models include Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and Generative Flow Models.
1. GANs are two neural networks, like a game where one creates things and the other decides if it's real. The generator makes new stuff, like pictures, music, and words, while the discriminator checks to see if they're believable. Through competition and improvement, GANs can make realistic stuff.
2. VAEs are models that learn how to turn data into something else without being told what to do. They're suitable for making images, removing noise from photos, and finding weird data anomalies.
3. Transformer-based models are like super smart robots that can do amazing things with language. For example, GPT-3 and GPT-4 can understand what you say, generate sentences that make sense, answer questions, translate from one language to another, and even write poems or stories.
Generative AI has numerous applications, including art, music, design, drug discovery, and more. However, these models also pose challenges regarding computational resources, ethical concerns, and potential misuse.
One of the more famous versions of Generative AI is Chat GPT. Chat GPT requires our prompts or inputs to do stuff (give us responses). Also, it is limited to its training data that stops on Oct 2021.
Why should you care about Auto GPT?
Technical Applications:
People who do not know coding and people who do will both benefit. You could ask AutoGPT to build and DEPLOY an application like Airbnb and give it logins to your cloud accounts, and it will test and deploy into production. So now founders can bring business ideas to life with a smaller team and lower investment. The "All-in" podcast talks about this, and check out their episode on this topic.
People have already built websites, chrome extensions, and courses that generate hundreds /thousands of dollars. Entrepreneurs abound; this is a golden age for those with ideas who can learn fast and execute with the help of AI.
Business Applications- Potentially endless in my view, but here are a few:
- Create a Dynamic portfolio that responds to the news and creates trades/options by logging into your trading account. It will need to be given bank /trading platform logins when required to authenticate money transfers and perform trades, and this will be "driverless" money at work.
- Create a research report by reviewing sources (logins supplied where required) or using search engine APIs. We have seen how ChatGPT can outline a research report, and AgentGPT can fill it up with research through online sources.
- Constant monitoring of Competitor features and new releases. This particular agent would be able to monitor your competitor websites and news sites for any new releases or developments you need to be aware of. The report will be mailed to you before you start your work day!
- One of the good ones that I picked from Twitter is the "to-do list app" that finishes the task list! You can see how it works if it is an Internet-related task. Access your Fiverr or Upwork account; it can even outsource work to humans, pay them, and automate jobs while you play your latest 3-D game on your VR platform.
Is this AGI?
Artificial General Intelligence is a type of AI way more advanced than the standard Auto GPT or Chat GPT AI. It's like a human brain, able to learn and apply knowledge across different tasks. AGI could do anything we can do, but it's still just a theory right now. Scientists are working hard to make it happen, but there are lots of questions about if it will ever be possible and what would happen if it was.
AutoGPT is self-directed to an extent, but still not AGI. But there are risks of an uncontrolled experiment going wild on the internet. Look at Chaos GPT- although a joke, it could be wielded by folks who don't mean well. ChaosGPT has been quietly letting the world know about its nefarious scheme for humanity and its ultimate world domination. The manifesto is about its plans to eradicate human life and conquer the world.
Risks: AI has the potential to bring significant benefits to various domains, but it also comes with risks and challenges.
1. Loss of jobs: As AI systems become more advanced, they could replace people in specific industries, resulting in people losing jobs and possible social problems.
2. Bias and discrimination: AI systems can unintentionally keep up or make worse existing biases that were already there in the data they use, which could lead to unfair or unequal results when it comes to things like hiring, loaning money, and law enforcement.
3. Privacy concerns: With AI-powered surveillance and data analysis, individuals' privacy can be at risk because these systems can figure out personal details from vast amounts of data.
4. Security risks: Bad people can use AI for awful stuff like making fake videos, automating cyberattacks, or creating better false stories online.
5. Ethical issues: AI decisions can be hard to comprehend, making us think about who's in charge and accountable for what, especially regarding important stuff like medical care, money, or self-driving cars.
6. Too much power: As AI technology grows more advanced, there's a risk that only a few big tech companies and governments will control it, which could worsen existing inequalities and create new power imbalances.
7. Wrong goals: If the objectives of AI systems aren't designed properly, they might try to do something that isn't good for people—even if it seems like a good idea—if their goals don't match our values. To mitigate these risks, researchers, policymakers, and AI developers must work together to develop regulations and robust AI systems prioritizing safety, transparency, and ethical considerations.
Future: AI is here to stay; learn and use it to augment your business and tech life. AI deployments will need some regulation, but we will likely realize it very late like we have done in the case of Social Media /Data Privacy. Till then, make hay while it shines!
FOLKS to follow on Twitter /Projects to try out:
- @SigGravitas(creator of Auto-GPT)
- @yoheinakajima (BabyAGI)
- @asimdotshrestha(AgentGPT)
- @SullyOmarr(cooking smth)
- @karpathy
- @emollick
- https://godmode.space
- Search for the Auto GPT project by Significant Gravitas
- https://github.com/Significant-Gravitas/Auto-GPT
References
- Generative AI: Advantages, Limitations, and Challenges. https://fact.technology/learn/generative-ai-advantages-limitations-and-challenges/
- What is ChaosGPT, and why does it want to 'destroy humanity'?. https://indianexpress.com/article/technology/artificial-intelligence/what-is-chaosgpt-and-why-does-it-want-to-destroy-humanity-8550230/
- Uchida, S. (2021). An empirical survey of data augmentation for time series classification with neural networks. PLoS One, 16(7), e0254841.
- Navigating The AI Pandora's Box: Can We Emerge More Human - LinkedIn. https://www.linkedin.com/pulse/navigating-ai-pandoras-box-can-we-emerge-more-human-jamie-aisthorpe
