“You have 18 months” The real deadline isn’t when AI outsmarts us — it’s when we stop using our own minds. www.theargumentmag.com/p/you-have-1... My concern is that this will not be true equally. short-thoughts.christopherlopes.com/blog/2025/re...

For many years I've desired to improve my vocabulary. I've implemented various techniques and routines. Alas, none has taken hold. My latest attempt has been to inform AI of my objective and to take this into consideration when generating responses. So far I'm encouraged by the resulting behavior.

Martin Fowler shares this thought from Rebecca Parsons: "...hallucinations aren’t a bug of LLMs, they are a feature. Indeed they are the feature. All an LLM does is produce hallucinations, it’s just that we find some of them useful." martinfowler.com/articles/202...

Why is it that we craft our work so AI can understand but we do not do so for people? It's like we have an expectation people should put forth the effort to understand, but AI has hard limitations, so we have to take extra steps for AI but can be lazy for others. www.heavybit.com/library/podc...

"But what they cannot do is maintain clear mental models. LLMs get endlessly confused: they assume the code they wrote actually works... This is exactly the opposite of what I am looking for." Insight on the difference of experience being reported zed.dev/blog/why-llm...

Growing up it became inculcated within me that education is about not knowing or understanding something. But what I try to remember is that education is about learning. One mindset produces fear and intimidation. The other exploration and insight.

When using the new study mode in ChatGPT and asking about typing gotchas in Python its reply was: "But for people used to strong, safe systems (like you), it can feel like trying to write a novel with crayons." That's one for the quote wall!

Trying out Google's Opal opal.withgoogle.com/landing/ Even with AI it's still missing the mark on low code experiences. It seem inevitable low code will be figured out, but that day is not today.

Sometimes I feel like this when trying to correct an AI response. I just can't stop until I convince the AI it is mistaken. xkcd.com/386/

Regarding AI agents: "They are basically a new programming control structure which can take English and point you to what to run next" Seth Juarez

I'm not having much success with the Gemini CLI. Using an API key I initially received too many requests errors. Now I'm receiving "The model is overloaded" errors. When I use Gemini within IDE it takes a long time for it to reply.

Agreed simonwillison.net/2025/Jun/27/... The practice of prompt engineering is important but the term has lacked rigor. Context engineering seems to better communicate the intricacies of crafting the input.

I now question if I need to bookmark webpages any longer. I spent a few minutes trying to find an article I recently read. No luck. I vaguely described it to o3. It thought for 1:34 and replied with the precise article.

Using coding AI agents is a skill. Agents in a loop can solve many problems where sometimes I find myself working with AI and being part of the loop, when an agent can have completed the task itself in a loop. Other times the agent needs human correction.

AI increases the set of people able to create software tools, but I fear most will not be be maintained. Previously it required a higher level of commitment to put something out there, thus weeding out those not committed to maintaining the tool. Now you can create without much commitment.

All of these instances of lawyers including AI generated mistakes makes me wonder about the thoroughness of their other work, a field which I thought was all about details, at least it seems that way based on whenever legal documents come my way.

Reports of AI usage likes this remind me of my experience learning math in school. I had the appearance of excelling but I didn't understand concepts, I could only apply procedures in context. Once I stepped outside of that artificial world I was lost. albertofortin.com/writing/codi...

"But there was also a lesson about the raw power of personality. Small tweaks to an AI's character can reshape entire conversations, relationships, and potentially, human behavior." www.oneusefulthing.org/p/personalit...

text-incubation.com/AI+code+is+l... "Software evolves more rapidly under the maintenance of its original creator, and in proportion to how recently it was written."

Humans like to create, thrive on it. In one sense AI enables more humans to create more things. On the other hand AI robs what many of us consider the part of the process which gives us joy in creation, leaving us the portions we consider drudgery. worksonmymachine.substack.com/p/the-coming...

What to make of normalizing "cheating". They're getting attention. So success? But what happens to society by labeling AI usage as cheating. To redefining what cheating means. The consequences of playing with morality is severe. cluely.com/manifesto

I'm extremely thrilled to see how quickly the semantic kernel team is adding support to new protocols from other orgs. First MCP and now also working on agent2agent. devblogs.microsoft.com/semantic-ker...

Add MCP Servers in .NET youtu.be/MKD-sCZWpZQ?... The nice bits about the video is not build the server itself, that's pretty standard, but how to use it in VS Code and publish it to docker. Everything is StdIO though. I'm waiting to say how MCP shakes out with SSE.

VS Code now has comparable AI features to other modern editors. youtu.be/dutyOc_cAEU?... It was only a matter of time. It's beneficial to me, but I welcome the editor competition, and so I hope it continues.

No one has a right or even a reasonable expectation to not be offended. We make progress not by bludgeoning our opponents into believing they are bigots and bad people, but by finding out what people think and why they think it. A Theory of Everyone, pg 325

Helpful example setting up a remote MCP server using SSE. We need more guides like this. blog.cloudflare.com/remote-model...

Exciting that both Microsoft and OpenAI seem to embracing Model Context Protocol. Unsurprisingly there's a lot of discussion on whether it's the best "protocol", but having a common "protocol" is welcomed by someone who dealt with browser differences for years. www.microsoft.com/en-us/micros...

Claude can write out its reasoning step-by-step. Does this explanation represent the actual steps it took to get to an answer, or is it sometimes fabricating a plausible argument for a foregone conclusion? Similar to Left-brain interpreter? www.anthropic.com/research/tra...

The stone soup comparison to how LLMs were developed is insightful[1], although it doesn't seem accurate. The stones didn't add anything to the soup while LLM researchers added and continue to add a lot of their novel content to the soup. 1: www.preposterousuniverse.com/podcast/2025...

I played the prompt security game Gandalf[1] after listening to a podcast with Mateo Rojas-Carulla[2]. I made it to the final level, but no further. It was a fun exercise causing me to think how prompts can be used in unexpected ways. 1: gandalf.lakera.ai/baseline 2: ainativedev.io/podcast/navi...

Don’t. Don’t ask for a yes. Instead, offer a chance to say no, but with a deadline. www.mooreds.com/wordpress/ar... There have been times when I've been excited to pursue a task and so have done this. It has worked out well. I'm thinking I should make it standard practice.

I was excited to hear that the Huggingface bonus unit for the Agent course was how to fine-tune a model to add function calling. But as I feared they went from wading into the shallow end to diving into the deep end. huggingface.co/learn/agents...

In research and academia, success often goes not to the one who understands first, but to the one who understands best. muratbuffalo.blogspot.com/2025/02/my-t...

Re Vance's speech and Amodei's statement on the Pair AI Action Summit, it's a tough spot. It would be great if policy races ahead with tech. But a benefit of democracy is that it's not easy to make changes. - gist.github.com/lmmx/b373b98... - www.anthropic.com/news/paris-a...

I enrolled in Huggingface's agent course. It's nice to learn about some of the underlying details I'm only sort of aware of. Thank you for the course @hf.co huggingface.co/learn/agents...

I hope these concluding words by Sam Altman is what occurs w/ AI There is a great deal of talent right now without the resources to fully express itself, and if we change that, the resulting creative output of the world will lead to tremendous benefits for us all blog.samaltman.com/three-observ...

Does this indicate AI intelligence is different than human intelligence? AI sees patterns we do not. Humans recognize patterns AIs do not. This seems to be a reason why it's important for humans to remain in the loop. The future is working together. chiraaggohel.com/posts/llms-e...

Reading the paper on DeepSeek it's interesting that examples were used to improve R1. I've been reading the book The Theory of Us. It discusses the incredible human learning by observing and mimicking others. It sounds similar to how R1 learned.

Value and Responsibility in AI Technologies conference at Gonzaga April 3 & 4. "Because of its pervasive import, it is more than timely that we should engage in reflective and critical thought on its ethical underpinnings." www.gonzaga.edu/informatics-...

"Language models can already write code better than most humans. This is going to bring the cost of building software down to zero." "These humans won’t write much code directly. Instead, they’ll manage teams of agents that build software for them." www.ycombinator.com/rfs#spring-2...

You can’t build real software without being precise about what you want. Every successful programming tool in history reflects this truth. AI briefly fooled us into thinking we could just chat our way to working software. danieldelaney.net/chat/

Will AI Agents, like OpenAI's Operator, replace robotic frameworks, like UI Path? It's a tough time to make a decision on where to invest. www.tessl.io/podcast/live...

If you value being trusting you will be seen as gullible If you value being flexible you will be seen as inconsistent If you value being stable you will be seen as boring Behavior makes sense from the actors perspective or the actor wouldn't do it www.hubermanlab.com/episode/dr-e...

I've been hearing more about Jevons paradox in relation to AI: As a resource becomes more efficient to use the cost of the resource drops, but overall demand increases causing total resource consumption to rise. en.wikipedia.org/wiki/Jevons_...

The intelligence of humans is tool use. We either slowly learn novel tools and how to use them or can quickly learn how to use them by being taught. The ultimate tool is understanding concepts and applying the concept across domains. We stand on the shoulders of giants.

Previously I’ve argued that progress towards ARC-AGI would result in higher reliability. The challenge with LLM agents is they need strong local domain steering to work reliability. Stronger generalization capability requires the ability to adapt to unseen situations. arcprize.org/blog/r1-zero...

'Oh my God, it's going to replace me.' But it's, like: Well actually, in fact, just like a lot of human jobs in the last centuries, what happens is most often is that the job gets replaced by a human using the technology, not by the technology itself. ~Reid Hoffman www.econtalk.org/why-ai-is-go...

An analogy to AI hard prompts is "view source" on web sites. One used to see and learn how something was accomplished on the website. With modern web frameworks that's no longer possible. Soft prompts are like modern web frameworks, more powerful but not as easy to learn from.

Fortitude presupposes vulnerability; without vulnerability there is no possibility of fortitude. An angel cannot be brave, because he is not vulnerable. To be brave actually means to be able to suffer injury. Because man is by nature vulnerable, he can be brave. alongthewayblog.net/2020/04/20/v...

Nice short write up on StructRag in .NET. So many things to try out! www.linkedin.com/pulse/struct...