Short-form insights you can read in under 5 minutes.
I recognize the serious risks posed by emerging technologies, particularly how they can deepen inequalities and spread misinformation. The push for fully autonomous systems raises real concerns for human safety. I believe the call for transparency, whistleblower protections, and a culture that embraces dissent is not only reasonable but essential for preventing blind spots in organizations. Regulatory bodies can help enforce these changes.
I recognize the serious risks posed by emerging technologies, particularly the perpetuation of inequality and misinformation. The rush towards fully autonomous systems raises real concerns for human safety. While the call for transparency and whistleblower protections is reasonable, changing an organization's culture to embrace dissent is challenging yet essential to avoid blindspots. Regulatory bodies can help enforce these changes.
I haven’t begun to think through all the ways my personal messages might be being used to train models (emails, family and friends social posts, messages…)
I know a great tool for summarizing all those changes in terms of use notices. Is that ironic? Or just sarcasm?
Ethical leadership derives from two values which exist in exquisite tension:
Integrity: Doing what you promise to do — as CS Lewis said, “…even when no one is watching.”
Empathy: Caring for the welfare of people not a projection of yourself. Engaging them on their own terms.
I love this idea. As with any resource intensive operation, part of sustainability is finding a way to consume resources off peak so that the provider doesn’t have to resort to the least efficient sources is supply. Given cloud services are often locally optimized, processing off business hours may make a difference.
Some transparency into the emergent behavior of these large models. Provides some insight into how the subtleties of how we write prompts activate different areas of the neural net and could elicite different responses.
In other words I’d suggest not letting our your inner domineering egotist out in your prompts. Unless you want the response inflected by decades of web discourse in response to domineering egotists.
A well-respected, thought leader told me, generative AI will create plenty of work in two years when autogen code starts failing in production. That will happen because it happens with human authored code. Another told me to think about the needs we could serve if custom software were less expensive to create. And that is true as well.
There will be a waste and harm to these tools because we are fallible, outcomes are uncertain, and our system is mercenary. Those are our givens. Even more reason then as software developers that we insist upon quality, learn how to use this tech more responsibly, and seek meaningful purpose in what we build with it.
I want leaders who care about what their employees think of them. Not because they want to impress or be liked or care about some of their coworkers but because they feel obligated to honor commitments made by their organization. They exist. Seek them out or become one yourself.
I frequently use large language models to enhance my coding and solutions. While I embrace the technology, I remain skeptical of the hype and the ethical dilemmas it presents. This study sheds light on the environmental costs of different models, emphasizing the importance of efficiency and responsible usage. I advocate for fine-tuning, smaller models, and thoughtful application to minimize energy consumption.
Providing access to open models and the processing required to develop applications with them for non-profits and non-commercial academic research is a great, pragmatic step towards positive impact.
This is exactly what I’m experiencing.
Agentic approaches might be powerful but they shouldn’t be thought of as a replacement for skilled problem specific engineering.
Think of it in terms of return on processing time. Invest in design time up front to avoid orders magnitude more processing expense at runtime
For many - if not all - reasons.
Here are some of my learnings using large language models to help automate cleanup on a large legacy code project:
Prompts must be very specific to the issue type being addressed and need to be carefully crafted. This is still a 1-to-n operation as the same issue type might appear hundreds or thousands of times in the code.
Avoid any extraneous noise in the prompt, instructions, and context. Context (whether to add full code, parent code, dependencies) is issue type-specific.
Use established solutions for solved problems, e.g. ctags for code tree traversal and diff/patch for integrating changes.
Model selection is important. Bigger isn’t necessarily better. Larger models can be more arbitrary and cost significantly more to run (whether you pay for it or the planet does).
Multi-agent systems aren’t magic. Applied thoughtfully they might act as a guard against outlier variances but each interaction needs to be thoughtfully crafted and wont make quality out of nothing.