Miguel García

Giving your AI a Job Interview As AI advice becomes more important, we are going to need to get better at assessing it

As AIs get smarter & more useful, our benchmarks become less useful. Measuring general knowledge or coding ability gives us only a glimpse into what an AI model can do.

Anyone who wants to use AI seriously for real work will need to assess it themselves. www.oneusefulthing.org/p/giving-you...

I'm interested in Linkedin!
By the way, I recently got a year of Perplexity pro for free, I think it was a Paypal deal or something like that.

If things get very extreme (AGI does everything and there are no jobs for most people), I guess the AI owners will have to share the wealth or face a violent revolution. Other option is that open source AI gets good enough that there are no monopolies.

I use only the Android app and I think i've never been logged out.

And 20 different invoices to manage instead of 1 big one... It's incredible how Facebook and Google (with other stuff) treat advertisers so badly and they still earn trillions thanks to them

Welcome to capitalism. It happened and will happen with every new tech revolution.

I hope cheap robotaxis (shared whenever possible) will solve a big part of this. Normal public transport doesnt really work outside of big cities.

Also, fewer invoices! (Both parties may save some time managing them)

You actually didn't. Customers could not care less if you are using that exact color in all your social media posts for example, only branding professionals with OCD care. Also, it's easy to apply your desired HEX color with AI, so if you take this color thing literally it doesn't make sense either.

This is a metaphor, it's not about colors or branding. It's about doing things manually to get the exact thing you want vs. doing it with AI and get a good enough result in 10x less time.

For most things, it doesnt really matter if you have the exact color you want or a similar one that looks decent. It's just wasting time in not important details. Same with apps and other stuff. As long as they work, you should probably use the most efficient tool (AI most of the time) and move on.

Cant read the whole thing with the paywall, but he always seems to assume that all this huge companies are only trying to advance AI in one direction (make bigger LLM models) and zero resources to other options. I dont have data (please share if you do), but i'm certain that this premise is false.

In 5 years basic AI will be almost free (it is already with open source models, especiallly smaller ones). No idea what frontier AI will be, maybe we will have AGI and most traditional SaaS will be dead.

Cost of AI APIs are falling sharply (something like 100x in 1-2 years) and most of these features dont even need the most intelligent/expensive models. Unless nobody pays for AI plans or many complain about AI features, I think is unlikely that most providers remove them.

Title is bullshit (and will get mentioned everywhere). The real fact is "about 5% of AI pilot programs achieve rapid revenue acceleration". But that's not the goal of most AI pilots. In the same article they say 67% of projects using third party tools succeed, but that wouldnt be viral...

That's not very precise. LLMs can totally give you sources (many different links). Also, compared to how many people get their info (tiktok, random websites and forums that google recommends...) I would say frontier LLMs with research features are much more reliable.

I guess it's a good comparison. I still hate the news feed though (always choose to see only "following" here & in threads, use newsletters and RSS to get informed...). They should keep the selector for power users.

Thanks! ✌️

Can you share a prompt/example? Thanks!

Fines for doing what exactly? By the way, I guess this is probably just clickbait, but they can probably fine you even if you dont live there, just for selling to EU customers.

You keep sending incomplete posts here...

Talking about X on Bluesky is a good way to get corrected too... 😬

The fact that the guy have 4 jobs and still delivers as expected means either he is a genius (probably not, if he was he could start his own thing, become really rich and retire) or that many companies expect very little from its workers (sad).

Do you think current AI tools are worse that other big sources of "information": fake news in social media, unreliable and manipulated search results by corporate interests, etc.? A good AI "filter" (maybe open source) could actually reduce misinformation...

Can you please stop retweeting (or whatever its called here) 10 posts praising your own article every time you write a new one? It pollutes our "following" feeds (nobody wants to see 10 times the same article) and adds no value. Thanks!

Do you have an opinion on n8n vs zapier, make & others? Are you doing only the n8n integration?

All the links? I only see 1 in the whole post. I'm not saying that he is lying, but without links we cant check the data, know if his sources are reliable...

Would be nice if he linked to sources. Otherwise we can only take his articles as personal opinions, speculations or even hallucinations (as with LLMs)...

Which tool do you use for the scraping? Do you automate some steps or just give the posts and get the URLs?

Can you give more details about the vibe prospecting part? 🤓