Columbus

The other day we were asked to take part in a accelerator program. As always some form to be filled out. I was short on time. In fact it was aleady past the official deadline. But the organizer wanted us absolutely in. So what did I do? I turned to ChatGPT to help me formulate the answers to the questionaire.

And now something scary happened.

One of the questions was about how our startup and product fit the challenge (see next screenshot)

I simply copied the questions into ChatGPT without any additional context. Here’s the answer I got.

So in fact without me providing the specific context of Squirro ChatGPT returned to me the description of another company’s answer to the same questions. I cross-checked this and now it gets scary: Above mentioned company submitted to that same challenge…

So ChatGPT reproduced an answer from somebody else – efficient caching, everything morphing into the same thing (ChatGPT producing the same answer regardless of who asks), the system knows who asked what, when…

PS: After a bit of prompt engineering ChatGPT produced a fairly good answer describing what we do instead what others do and have submitted to the challenge.

We released GPT for the Enterprise – https://squirro.com/enterprise-generative-ai-and-large-language-models/… Obviously we’re trying it on our own stuff, e.g. our ISO Rulebook. And as obvious we try to see if we can break it. Here’s a test. We failed… You can try it yourself: https://start.squirro.com

“Since AI has been around for many years already, I expect a comparable diffusion in one or two years.”

We spoke about this in the EM Interview in February 2022.

https://link.springer.com/article/10.1007/s12525-021-00516-w

Adopting large language models (LLMs) is not without its challenges, especially for enterprises.. To overcome these challenges and achieve maximum benefit from these models, the combination of composite AI with large language models is the best way forward for enterprises. 

Better Control and Customization:

One of the main benefits of combining composite AI with large language models is the ability to have better control and customization over the models. Composite AI allows enterprises to combine multiple AI models to create a custom solution that fits their specific needs. This is particularly important for large language models, which can be too generic and may not provide the level of control that enterprises require. 

Improved Accuracy and Performance:

Another benefit of the combination of composite AI with large language models is improved accuracy and performance. Large language models can generate huge amounts of data, which can be difficult to manage and interpret. Composite AI allows enterprises to use multiple models to analyze and interpret the data generated by large language models, leading to improved accuracy and performance. This is particularly important for applications such as customer service, where the accuracy of the model’s responses can have a significant impact on customer satisfaction.

Better Data Privacy and Security:

Data privacy and security are major concerns for enterprises when it comes to adopting large language models. Composite AI allows enterprises to control the data that is used by the models and ensures that sensitive information is properly secured. This can help to mitigate the risks associated with large language models and ensure that enterprises can adopt these models with confidence.

Lower Costs:

Adopting large language models can be expensive, both in terms of hardware and software costs. By combining composite AI with large language models, enterprises can reduce these costs and achieve more cost-effective solutions. This is because composite AI allows enterprises to use multiple models, each of which can be optimized for specific tasks, reducing the overall cost of the solution.

Better Integration with Existing Systems:

Large language models can generate huge amounts of data, which can be difficult to manage and integrate with existing systems. Composite AI allows enterprises to integrate multiple models and manage the data generated by large language models more effectively. This leads to better integration with existing systems and ensures that the data generated by the models is properly stored and managed.

In conclusion, the combination of composite AI with large language models is the best way forward for enterprises adopting large language models. By adopting this approach, enterprises can maximize the benefits of large language models and ensure that they are delivering the results they need.

PS: This is btw the opinion of ChatGPT itself

Adopting large language models (LLM) such as ChatGPT  in an enterprise is not without its challenges. In this post, we’ll discuss some of the key challenges that enterprises face when it comes to adopting large language models and how they can overcome them.

Data Privacy and Security Concerns:

One of the biggest challenges that enterprises face when adopting large language models is data privacy and security. These models are trained on massive amounts of public data. For a LLM to become useful in the enterprise context they need to be retrained on often sensitive information such as personal data, financial information, and confidential business information. To mitigate these concerns, enterprises need to ensure that their data is properly secured and that the models are not accessing or using sensitive information without permission. This requires implementing robust security measures such as encryption, data masking, and access controls.

Integration with Existing Systems:

Another challenge that enterprises face when adopting LLMs is integration with existing systems. Large language models can generate huge amounts of data, which can be difficult to manage and integrate with existing systems. Enterprises need to ensure that the data generated by the models is properly stored and managed, and that it can be easily accessed and integrated with existing systems such as databases and analytics platforms.

Cost:

Large language models can be very expensive, both in terms of hardware and software costs. Enterprises need to ensure that they have the budget to purchase and maintain these models, as well as the infrastructure to support them. This can be a significant challenge, especially for small to medium-sized enterprises.

Skills Shortage:

Another challenge that enterprises face when adopting large language models is a skills shortage. There is a lack of talent with expertise in these models, which can make it difficult to implement and use them effectively. Enterprises need to invest in training and development programs to ensure that their teams have the necessary skills to use these models effectively.

Bias and Halluzination:

Large language models can be biased due to the data they are trained on, which can lead to incorrect results. Enterprises need to ensure that their models are trained on unbiased data and that the predicted results from the LLM are corroborated against actual data in the enterprise. 

In conclusion, while large language models offer significant potential benefits to enterprises, there are several challenges that need to be overcome in order to adopt them effectively.

ChatGPT is one of the most advanced language models in the world. While it is often referred to as a “AI language model,” it is important to note that it is not truly autonomous or intelligent in the same way that humans are. Instead, ChatGPT can be thought of as a highly sophisticated “stochastic parrot.”

A parrot is a bird that has the ability to imitate sounds and mimic speech. While a parrot may seem like it is speaking of its own accord, it is actually just repeating what it has heard before. Similarly, ChatGPT is a model that has been trained on a large dataset of text and has learned to generate responses based on the patterns it has seen in that data.

However, unlike a parrot, ChatGPT uses probabilistic methods to generate its responses. This means that instead of simply repeating what it has seen before, it generates new responses by predicting what is most likely to come next based on the input it receives. This is why ChatGPT is referred to as a “stochastic” model – it generates responses based on probability rather than determinism.

While ChatGPT is extremely advanced and can generate responses that are very human-like, it is still limited by the data it was trained on. For example, if the data it was trained on contains biases or inaccuracies, ChatGPT will also generate biased or inaccurate responses. Additionally, since ChatGPT has not had direct experiences in the world like humans have, it can sometimes produce responses that are nonsensical or that lack context.

Despite these limitations, ChatGPT is still a valuable tool for a wide range of applications. For example, it can be used to generate natural language responses in customer service interactions, to help with content creation, or to assist with language translation. However, it is important to keep in mind that ChatGPT is not a replacement for human intelligence – it is simply a tool that can be used to augment human capabilities.

In conclusion, ChatGPT can be thought of as a highly sophisticated “stochastic parrot.” While it is an impressive model that can generate human-like responses, it is limited by the data it was trained on and is not truly autonomous or intelligent. However, it remains a valuable tool for a wide range of applications and has the potential to greatly augment human capabilities.

PS: The text was generated by ChatGPT….

Robert Solow, an economist, said in 1987 that computers show up everywhere except for the productivity statistics. It took a good two decades at the time for this to fundamentally change.

I predict the same “Solow Paradox” will apply to AI and its newest incarnation LLMs (of which GPT is a variant). Over the next few years a lot will happen in terms of adoption that will not show up positively in productivity statistics (in some cases it will actually show up negatively… 😉). 

The difference will be that gains will start to show up at the end of a single decade of adoption instead of two.

In the finance industry a fat finger event describes a typo with consequences. A few years ago, a Deutsche Bank trader mistakenly transferred 6 billion.

Prediction: This is exactly what will happen with GPT as well. Actually, it will happen with system like Github Copilot. Some programmer isn’t paying attention, doesn’t check the code fragment in which – can also happen GPT – the wrong variable is in it and the live system executes the equivalent of a fat finger..

With the release of ChatGPT, a wide audience has become aware of what AI can do today.

I can remember when we started using the early versions of Google for the first time in 1997/8, we knew: That was it for AltaVista & Co.

The advances in Large Language Models (LLMs) of which GPT is one are the same: a moment when the future will be different from the past.

2022 was truly a turning point year.

PS: We as citizens should regulate AI quickly and comprehensively as I suggested for some time now. I compare it to an airplane: it’s only safe because it’s strictly regulated. We should apply the same logic to AI.

Inflation is a nuisance and yet an opportunity for some. However, it is more of a devilish opportunity: inflation is caused by profiteering.

It is caused when companies or individuals increase prices for their own benefit instead of reflecting changes in supply and demand. One such case might be the Coop carrot salad.

Until last autumn it cost 2.80 CHF. Since the New Year, the same salad has cost CHF 2.95. That is a price increase of 5.3% and well above the general inflation rate of 3% here in Switzerland… Coop, can you explain that above?