Eric @ SkillFoundry.io

Eric @ SkillFoundry.io

@skillfoundryio.bsky.social

Software architect, entrepreneur, teacher, gamer, father, not in that order! My opinions are that of my employer, because it’s my company. Follow me on YouTube for developer content for beginners and early career. https://youtube.com/@skillfoundryio

105 followers163 following223 posts

Seeing Anthropic and Google complain about distillation is particularly delicious because those Chinese firms paid for API access, unlike certain companies that just looted the internet. Let me break out the world’s smallest violin. Your product has no moat.

Crafting good assessments and feedback rubrics are arguably the most important and challenging parts of the learning experience. The availability of AI tools has only increased the challenge. What are your plans for proper assessment going forward?

I would have to be pretty desperate for work to show up for an interview and find an AI avatar waiting for me and not just immediately close the session. I wonder how many professionals with in-demand skills and employment options would do the same?

Funny but not funny. If the vibe coders took 6 months learning to code, they’d be orders of magnitude more productive. Even if they luck their way to a viable product, someone with actual skill and the same tools can just make a better version. Ask AI about second mover advantage.

A little bit dark, but I have to ask people who live in high wage/high cost of living countries: If all you're doing is prompting an AI tool to get all the work done, what makes you think anyone will pay you a living wage to do so? You're adding zero value to the process.

A lot of the time when I first get called into consult on learning experiences I find either a text wall or a video "information dump" that serves as training. We can do so much more for our learners. If information dumping all learners need, then the completion rate of MOOCs wouldn't be under 5%.

Maybe I'm being overly critical, but if you're doing a remote interview: - Have your camera on. - Make sure your space is clear of clutter. - Make sure there isn't a lot of background noise. Some candidates really aren't thinking about how they present themselves.

I wonder how much money some decision makers could earn if they blocked off a few hours of their time a month and let people bid on booking a meeting with them. I'd be more than happy to listen to any sales pitch for the right price... What do you think?

If you're worried about AI tools replacing you as a developer, just go read the GitHub issues for some of them. "Doesn't work" "OMG it's burning so many tokens failing, stop charging me!" "It fails when the project gets 'large'" (they think large is 30 files) etc.

One of the most infuriating things about AI is what companies aren't doing. Client sends me an email with some days and times to meet. How about you highlight the ones where I'm available and pre-fill a meeting invite for me? #WhoTheHellIsTheProductManager

Watching Microsoft and OpenAI shifting their message to "ease up on the hype train" is vindicating for me. Also, watching OpenAI launch search was interesting as I've been warning AI SAAS products that they don't have a good moat. (Looking at you, Perplexity). Happy Monday! Be safe out there!

One thing that stands out to me when I'm advising on tech hiring is how few people in the hiring process have been trained. I think most leaders agree that having the right people on staff is important, so why is there a lack of investment in training the people doing the hiring?

I've been hearing rumblings of "leetcode" interviews starting to go away. I predicted this a while ago, and it's interesting. If AI tools can solve these problems without actual reasoning, then why did so many people think they were good selection criteria? (I've never used them in my hiring)

At the time of this writing, the training data for Claude is from April 2024. One of the downsides of AI tools in tech, especially in things like cyber security, is that they aren't current. It's important to know the training cutoff when you use a model!

One of the current techniques to improve accuracy is to have multiple AI agents work on the same task. If they agree enough, we assume it's good enough to go to the next stage. Just remember, the consensus of these tools was that there are 2 r's in strawberry. Consensus <> Correctness

Anyone else confused how Altman can claim “PhD level reasoning” and also claim AI can do 5% of all tasks in the economy today. I mean, I know we joke about how those who can’t do, teach, but I’m pretty sure the average PhD can do more than 5% of tasks.

Why couldn't Google create a super bowl ad for their AI product without hallucinations in it? What chance does an SMB have of getting it right? Why would anyone attempt to use these tools without supervision for anything that requires precision? Why are the Kansas City Chiefs so overrated?

Remember that movie, The Butterfly Effect? That’s basically how I explain to non technical people why agentic AI won’t work out the way they hope. If any part of the chain hallucinates or comes to an incorrect outcome that error will compound further down the line.

Being a founder is very lonely. No one cares the way you do. No one will work as hard as you on your business (rightfully so, but still). No one sees the extra hours you put in. Do you know a founder? Give them a hug. Show genuine interest in what they're doing. It's always appreciated!

When engaging in deep learning, you should have notes and a journal. Notes are your documentation, organized by topic, with details, and serve as quick references. The "what". Journals log your learning narrative. Insights, progress tracking, struggles, and achievements. The "why and how".