According to Apple Photos AI, I'm a French Bulldog 🐶

As a Chief Technology Officer, I am bullish on generative AI. Yet, I also recognize the potential problems brings, such as biases and inaccuracies.

It seems like people either see generative AI as an amazing tool that we should fully embrace or as something harmful that we should completely avoid.

I believe we need a balanced approach. We should understand what generative AI can do and what it cannot do. It's important to be aware of its limitations and to set up safeguards around it.

For example, today I took a selfie in the swimming pool, and the Apple Photos AI identified me as a French Bulldog. While I laughed at it, it highlights a serious issue.

If an AI system can’t identify something correctly, it will lead to wrong decisions, especially if people rely on it without any human checks.

That’s why, when we implement any AI system, we need to have checks and balances in place. There should be a clear workflow that includes multiple levels of review, starting from a first-level review to an approver and so on.

We can’t eliminate all human oversight, but AI can definitely enhance our existing systems. It's crucial to have manual reviews to ensure everything works as it should.

#gptusage

Importance of Context while using Big Data

There is a catchy phrase in computer science, "garbage in, garbage out". Usually it is used with respect to data. It is equally applicable for setting context while querying data or using GenAI.

I used to work for the Ministry of Corporate Affairs in India, which keeps a registry of all registered companies in the country, among all the other things. For a report at an upcoming conference, the then minister requested total number of companies in India. The person responsible for reporting ran a query, asking for all the companies in the database. They found a number, something like 5,322 companies, and reported to the minister.

During the conference, the minister confidently stated that there were 5,322 active companies in India. The audience pointed out that the number was incorrect. Many of those companies had been struck off the registry or were in the process of striking-off and then there were inactive companies. The actual number of active registered companies was much lower. None of us foresaw the confusion that a simple number could cause. It was only a minor detail in so many other things that were discussed that day.

I've seen mistakes like this happen with GenAI tools since I've been using cursor.ai. If you don't set the context properly, these tools will generate ineffective, buggy code.

Sometimes Claude (or cursor) would generate class components instead of functional ones, or react code instead of preact code.
GenAI tools can hallucinate, but before you blame the tool, maybe check if you set the context wrong.

#gptusage

Setting up android development tools on Mac (either react-native based or flutter based) is so damn hard.

Be it go or python or even front-end frameworks like #preactjs or #vuejs , they are easy to get started (on Mac).

Getting started with mobile development is so hard on Mac. Why?

I followed instructions on https://reactnative.dev/ / https://docs.flutter.dev/get-started/install still no luck.

I even tried Claude. Nothing worked.

The last error I got related to too many files open. Tried everything in this: https://github.com/facebook/watchman/issues/923 still no luck.

I have developed apps for Palm OS. It wasn't that hard to develop an app and install on your device, even though it was around two decades back.

#devnote #gptusage

Is this the beginning of GenAI crushing enterprise SAAS?

Klarna announced that they are discontinuing Salesforce and Workday, replacing them with an internally developed system using Generative AI code generation tools. This news is gaining traction, with many saying that Generative AI will soon replace enterprise software.

First, the actual impact on the company's sales will only become clear after a few quarters or even a year. If the new system negatively affects sales, the company might revert to traditional SaaS solutions.

Second, this move will likely lead to a trend where Chief Marketing Officers (CMOs) and Chief Sales Officers (CSOs) turn to Generative AI to create their own tools. Internal IT departments often face compliance and priority issues, which can be a roadblock. With Generative AI, CMOs and CSOs might develop their own applications, bypassing these hurdles.

Third, due to this shift, a new market might emerge for maintaining software applications created by Generative AI. This could open up a range of new opportunities and services.

What are your opinions on this?

#gptusage

Will Generative AI Lower Software Development Costs?

With all the advancements in generative AI, a key question arises: will it reduce the cost of software development? Specifically, will the costs associated with enterprise software in B2B and B2C sectors go down? I don't think so, and I have two main reasons for this.

First, I recall the time when banks and airlines were transitioning to paperless operations. Before this shift, these companies relied heavily on paper for their processes. For instance, banks used to send statements to customers in paper form, and many manual processes were conducted on paper. When they moved to digital systems, there was an expectation that this would lead to cost reductions, which would then be passed on to customers. People expected banks would offer higher interest rates and airline prices would drop. However, that did not happen. The main reason for this is human greed; companies often prioritize profits over passing savings to customers.

Secondly, when we look at cost allocations, we see that expenses are simply shifting from one budget to another. Instead of spending on employee costs, companies will now invest in tools and infrastructure to run AI models and large language models (LLMs). While they might save on employee training, there will still be significant costs involved in implementing and maintaining these AI tools. Generative AI and the tools that create code are not free solutions, and we need to recognize that.

What are your thoughts on this?

Capture your thoughts into valuable notes using @thoughttonote

#gptusage

Amazon introduces a #chatbot called Rufus. It is supposed to help shoppers in answering questions about the product.

But it is a wrapper over a gpt. So you can ask programming questions too ;-)

They didn't incorporate the guardrails that they talk about in every event ;-)

src: https://x.com/jjude/status/1834214063720595544

Though they seemed to have fixed it now

#gptusage

GenAI solutions are often presented as “cost saving” solutions to CXOs.

• it can replace your developers
• it can replace your marketing team
• …

The more I deal with AI solutions, they are never cost saving if you take a full perspective. It is taking budgeting from one bucket to another.

There is cost of infra, continuous fine-tuning etc.

What I am thinking now: it makes you do more for the investment. You will probably spend more overall but you will also do more.

#gptusage

I uploaded a screenshot of a login page I liked to #claudeai and asked to write code.

Act as a UI designer. Look at the enclosed image and design the same in plain HTML and CSS. Use Tachyons css for styling. Pick a Google font that resembles the font in the image.

The generated code is not perfect but it is close to the screenshot. It is becoming more and more easy to write code with Claude.

I might use this login page design for @thoughttonote

#devnote #gptusage

Been reading Walmart Tech blog on how they are using AI. These are some of the insights I got:

Search is where they are using AI more.

A customer can simply key in their need in natural language: “Help me plan a football watch party” or “What supplies do I need for a newborn?” and voila! Walmart’s GenAI Search gets to work pulling the list of all the relevant products

What is interesting is that,

our system will create distinct categories that collectively cover all possible customer needs, avoiding overlap or gaps

They are also using for product summarization and product comparisons. Their product comparisons seems interesting. It is not just on features, but

fulfillment, price, unique specifications, customer ratings

They utilize voice too.

pair their Walmart accounts to their smart speakers and mobile devices

It is not just for customers. They use AI for their in-store agents too:

Imagine you want to pick up a last-minute gift for your friend, but the Walmart store you walk into does not have the item you want in stock. Not to worry, because our associates can use real-time inventory data to find a store nearby that does have it in stock and reserve it for you, with the option to pick it up or have it delivered to your doorstep.

In their recent earning call, the CEO said:

the GenAI product that we're using has helped us populate the attributes and the characteristics of hundreds of millions of items and that would have taken, as I said, 100 times longer if we had
tried to do that manually

All of this would need tons of data, possible combination of products to an intent. They have invested heavily in data & ML. They have their own LLMs custom built for their usage.

#gptusage #ecommerce #retail

Creativity is made, not generated.

...Built on a foundation of theft, the technology is steering us toward a barren future....

src: https://procreate.com/ai

First CEO, as far as I know, who is taking a stand against AI.

Will he be an outlier? Or will others follow him in releasing "AI free" products?

#gptusage #ai