Tag Archives: AI

Bias in LLMs

What are the potentials for biasing LLM models? We dive into biases both in good ways and in bad ways.

Is the expertise that we’re feeding into these models is not sufficient to actually drive the outcomes that we’re looking for? We’re going to be eliminating humans out of the loop in a relatively short period of time. Both outcomes, at the moment, feel equally probable, which is troubling.

We dive into how and why that happens, what’s going on, and some concrete tips for how you can improve your prompting to avoid these same pitfalls.

Transcript: otter.ai/u/v3MaWiCWEe-G1ar2O0…?utm_source=copy_url
Photo by Marta Nogueira: www.pexels.com/photo/pink-and-bl…or-text-17151677/

Rob’s Hot Take:

In the Cloud 2030 Podcast episode from September 7th, Rob Hirschfeld explores the topic of bias in large language models, emphasizing the ease with which the output and tone of these models can be influenced by initiating them with different idiomatic English dialects. By demonstrating that variations in greetings like “bonjour” or “howdy” yield distinct results, Hirschfeld underscores the importance of crafting prompts and setting the right tone to unlock the embedded expertise within the models. The conversation delves into the fascinating and somewhat alarming aspects of bias in large language models, offering insights that encourage listeners to engage with the full discussion. Those interested in participating in ongoing conversations can find more information about Cloud 2030 at the2030.cloud.

CoDev With LLMs?

Can large language models effectively supplant developers and DevOps engineers?

Today we go deeper into how the models can be trained, if they can be trusted, and what is the upside or positive use case in which we really turn LLMs into the type of weighing person experts that they have the potential to be versus simply something that turns up the volume on how fast you generate code.

We also talked about the downsides of that type of model and the potential upsides of how powerful using these tools as assistants could emerge to be as a key aspect here to transform and improve the outcome for work.

Transcript: otter.ai/u/u3bArfIvx40oUXnRLt…?utm_source=copy_url
Photo by Pixabay: www.pexels.com/photo/aeroplane-a…e-aviation-33224/

Rob’s Hot Take:

In the Cloud 2030 Podcast episode from August 31st, Rob Hirschfeld discusses the potential of using large language models to enhance DevOps and development outcomes. The conversation emphasizes the possibilities of leveraging AI to improve codebases, facilitate refactoring, and encourage code reuse by tapping into the knowledge embedded in existing code bases. Hirschfeld envisions a future where AI assists developers in reducing technical debt, maintaining code more efficiently, and consolidating code intelligently, ultimately leading to improved development practices. The episode explores the challenges and investments required for realizing these outcomes, encouraging listeners to delve into the full podcast for a comprehensive understanding. To engage in further discussions, interested individuals can explore the Cloud 2030 podcasts and join the conversations at the2030.cloud.

Can we regulate LLMs? Should we?

How do you regulate large language models? We look at the challenges of regulating these AI approaches and how governments and companies can approach it. We untangle how these models work, and dive into the mechanics of what information is controllable. We walk through concrete information that is a benefit to you here as our listener, and incentive for you to join us in future conversations as we continue to unravel it.

In addition, John Willis was on the panel today, and he started us off with a story about API’s, Amazon, Jeff Bezos, and O’Reilly from the warmup. So you’ll get a short bonus story by John Willis before we start.

References
ised-isde.canada.ca/site/innovation…panion-document
www.europarl.europa.eu/news/en/headl…xt=Parliament‘s%20 priority%20is%20to%20make,automation%2C%20to%20prevent%20harmful%20outcomes
www.trade.gov/market-intelligenc…i-regulations-2023
content.naic.org/cipr-topics/arti…ial-intelligence

Transcript: otter.ai/u/dBRQBFNz8d01taQ-iM…?utm_source=copy_url
Image: www.pexels.com/photo/measuring-g…tar-pick-3988555/

Rob’s Hot Take:

In a Cloud 2030 podcast episode, Rob Hirschfeld, CEO and co-founder of RackN, discussed the complexities of regulating large language models. He highlighted the stark differences in approaches between the US, focusing on model risks, and the EU, emphasizing user rights protection. Hirschfeld expressed concern about reconciling these varying perspectives, especially regarding data rights preservation and understanding the risks associated with using such models, particularly when algorithms cannot be fully validated. He invited listeners to engage in the ongoing conversation at the 2030.cloud.

Data Darkages – do LLMs drive paywalls?

A coming Data Darkage is on its way, where we’re watching Reddit, Twitter and other companies take what used to be publicly available information and put it behind a paywall or gate.

Because of the way large language models are using this data and the value of the data, we are expecting to see that trend accelerate. This will have profound implications for how we think of, share, and use data in the coming years.

Transcript: otter.ai/u/e1XCyhSa9V81bgMpbo…?utm_source=copy_url
Photo by Pollianna Bonnett: www.pexels.com/photo/young-brune…e-chair-17687131/

AI And Technical Debt

We dig into a topic written about by Eric Norlin or SK ventures about technical debt and AI. In this episode, we discuss the consequences of generative AI could be radically transforming the way in which we generate code and deal with code that has been generated in technical debt.

We explore some fascinating concepts about how fast we can iterate, how we change the dynamics of building software, building automation, and the expertise required to architect systems. This leads pretty far down in the path towards disruptive thinking, and how this could reshape the entire industry.

Source: skventures.substack.com/p/societys-te…and-softwares
Transcript: otter.ai/u/MEtVkoNnZeCu0JHa30…?utm_source=copy_url
Image: www.pexels.com/photo/piggy-bank-…a-flower-4886900/

Rob’s Hot Take:

In a discussion on the Cloud 2030 podcast, CEO and co-founder of RackN, Rob Hirschfeld, highlighted the changing landscape of expertise in emerging technologies like AI. With the cost to build and iterate dropping significantly, expertise is no longer primarily applied during the building process, but integrated into design and testing sequences. The advent of generative AI has the potential to revolutionize how we design and build automation, software code, and technical systems, necessitating a redefinition of expertise in this rapidly evolving field.

AI Time To Decision

We talk about improving the time it takes to make decisions – called time to decision, a topic that we like to address quite a bit. We started with the news of the day around AI, ml Chaffee GP, and learning models.

We asked ourselves if AI/ML and generative AI could change the way expertise is used to make decisions and improve the time to the decision for experts. What type of implications would that have in the market?

If you’ve been tracking this subject, I know you will find this exciting and interesting.

Transcript: otter.ai/u/L3Kb_I4fe0ggAr9nDE…?utm_source=copy_url
Image: www.pexels.com/photo/person-touc…rm-clock-1198264/

Data Gravity vs AI and Metadata

We check in on data gravity to see how generative AI and conversations about metadata and thinking on data lakes impacts data gravity thinking in general.

Data gravity is a concept that has been propagated by David McCrory, a friend of mine, who defined this idea that data itself, the aggregation of data, the use and transit of data has a gravitational effect. That it pulls more data to it as well as workloads towards it.

We jumped right into impacts of data gravity in this conversation.

Transcript: otter.ai/u/qKf75W8OvZHkMtVQRM…?utm_source=copy_url
Image: www.pexels.com/photo/action-anim…co-bucking-33251/

Digital Twins + AI = WOW

How can the intersection of generative AI machine learning and artificial intelligence be applied to environments using digital twins? Today we discuss digital twins and artificial intelligence.

How can we improve the simulations, the systems, the interactions that we build? How can we correctly model complex components of everything from cars to pumps in ways that allow us to then build on top and build more intelligent systems.

We come up with some grounded examples.

Mentioned: projectarrow.ca/
Transcript: otter.ai/u/A79s08jpJ4-UPyT313…?utm_source=copy_url
Image:www.pexels.com/photo/two-bernese…on-floor-9040438/

Rob’s Hot Take:

In the Cloud 2030 podcast episode on digital twins and AI, Rob Hirschfeld discusses the potential of using digital twins in handling real-world disasters, citing the recent train derailment in Ohio as an example. The concept involves quickly creating a digital twin of a disaster space to enable robots to learn, adapt, and efficiently mitigate the situation. Hirschfeld emphasizes the unprecedented opportunities for improving environmental interactions, responding to crises, and highlights the sophistication of ideas discussed in the episode. He encourages listeners to explore the full conversation on digital twins and AI at the2030.cloud.

Meta Data, Dark Data and Intent

https://soundcloud.com/user-410091210/meta-data-dark-data-and-intent

Data comes from many different places, sources, and ways. Some data we call dark data, which is data not accessible to you, and all of it relevant. Today we talk about metadata as part of the governance control management exposure of data.

An important layer beyond the data itself is the governance intent, how people access it, and how you combine data. We discuss exactly what that is, but still only touch the surface.

References:
yago-knowledge.org
en.wikipedia.org/wiki/Ontology

Transcript: otter.ai/u/rP_4gi-iDont0r2CaZ…?utm_source=copy_url
Image: www.pexels.com/photo/apples-and-…n-books-14098062/

Rob’s Hot Take:

In the Cloud 2030 Podcast episode on metadata and building a data control plane, Rob Hirschfeld emphasizes the challenge of controlling data egress due to its diverse sources and destinations. He contends that rather than attempting to create a locked box for all data, the focus should be on embedding information, particularly metadata, to control data consumption. Hirschfeld envisions a distributed system for a data control plane, involving multiple parties managing and providing consistent rules for data use, acknowledging the complexity of data storage across various locations. He encourages listeners to explore the insightful February 2nd episode and engage in ongoing conversations at the2030.cloud.

Metadata For A Data Control Plane

https://soundcloud.com/user-410091210/metadata-for-a-data-control-plane

Metadata is information that travels with the raw data that provides context, provenance, security, authorship, controls, and indexing.

The number of ways that you can expand the use of data is controlled by adding metadata. It creates a change in how we look at and manage data. Instead of creating control systems that contain the data, it’s actually packaged the control infrastructure, or the data control plane, as part of the data so that all of the systems can participate in it.

We also talk about data mesh a lot in that context.

Transcript: otter.ai/u/bho2SHu0N1YfECix6X8qNvKTlB4
Image: www.pexels.com/photo/pensive-eth…up-phone-3760607/

Rob’s Hot Take:

In the Cloud 2030 Podcast episode from January 19th, Rob Hirschfeld discusses the challenges and opportunities associated with metadata, particularly in the context of managing and sharing data effectively. The conversation explores the concept of packaging raw data in a way that makes it available on request without being included, accompanied by metadata detailing its provenance, access permissions, context, and even expiration dates. Hirschfeld emphasizes the importance of building a data control plane to navigate the complexities of data consumption and production, envisioning a future where metadata plays a crucial role in weaving together diverse data sources. Listeners are encouraged to explore the full 40-minute discussion for comprehensive insights, and to join ongoing conversations at the2030.cloud.