Paul-Mehdy

Partner Data & AI at

[Paul-Mehdy @PaulMehdy]

https://t.co/PTy6PoTbom

[Paul-Mehdy @PaulMehdy]

Too often, we fixate on the AI model when the real key is how we define the problem. Take predicting tomorrow’s weather: you could use binary classification for ‘good or bad,’ multi-class for ‘sunny, cloudy, rainy, snowy,’ or three regressions—temperature in Celsius, precipitation in mm, and humidity in %. The approach matters more than the algorithm.

[Paul-Mehdy @PaulMehdy]

An 'LLM Trends' like Google Trends would be super insightful into what humanity's all about.

[Bindu Reddy @bindureddy]

People actually think an LLM can generate a trading bot that will make them millions.

The number of people who ask ChatLLM to develop one and get annoyed and email us, is pretty astounding 🤯

No wonder so many people are impatient for super intelligence! They believe they can become rich over nite 🤣🤣

[Paul-Mehdy @PaulMehdy]

LLMs reflecting human biases is a reminder of our responsibility to implement rigorous guardrails and input sanitization for ethical AI development.

[Jeremy Nguyen ✍🏼 🚢 @JeremyNguyenPhD]

@emollick Didn't realise it was a formula at the time—I recently published a paper in this genre.

In this case, anchoring bias affects LLMs: mentioning an arbitrary high or low number affects subsequent forecasts, even if you ask to ignore previous instructions or use chain of thought.

[Paul-Mehdy @PaulMehdy]

🔚 In conclusion

A balanced evaluation framework is key to find that sweet spot where accuracy meets practicality and let the business thrive!

10 of 10

[Paul-Mehdy @PaulMehdy]

Building the best model: Balancing competing trade-offs

Building a high-performance machine learning model often centers on accuracy, but neglecting other important factors can lead to long-term issues.

Thread 🧵

1 of 10

[Paul-Mehdy @PaulMehdy]

↳ The framework itself allows you to compare and choose easily between models.
↳ "Good enough" avoids over-investing in iterations or features, ensuring effective resource allocation and balancing perfection with realistic resources.

9 of 10

[Paul-Mehdy @PaulMehdy]

By implementing effective monitoring strategies across these areas, you can ensure that your AI models continue to deliver value and support your business objectives over time.

12 of 12

[Paul-Mehdy @PaulMehdy]

Monitoring AI models: Best practices for business success

Monitoring AI models is crucial to:
↳ Maintain business continuity
↳ Preserve customer trust
↳ Optimize costs

Focus on five key aspects for effective AI model monitoring:

Thread 🧵

1 of 12

[Paul-Mehdy @PaulMehdy]

Evaluate whether spikes in usage align with your business objectives to ensure that your resources are being utilized efficiently.

11 of 12