Alex

Interactive AI Tool Builder | Creator of

[Alex @AlexZheLin]

You ask an LLM to do an analysis of something for you. What do you next?

Try to understand the output? Ask follow up questions? Verify the accuracy?

The work of Analysis isn't dead, it's just entered a new era

[Alex @AlexZheLin]

@themgmtconsult Hiring consultants for frontier work is not a model I can say I frequently encountered.

I've seen this done more by internal research departments or buying out start ups doing frontier stuff.

[Alex @AlexZheLin]

@themgmtconsult Yeah, previous engagements with impressive sounding clients to show "you've done it before"

[Alex @AlexZheLin]

@themgmtconsult Then again, I think people love to be told that there is an easy and fun solution to their problems.

"Just stop thinking and only do things your gut tells you to want and you will have success. The problem is you are constraining yourself"

People just love this stuff.

[Alex @AlexZheLin]

How often do pick an option that is not (Recommended) anon?

[Alex @AlexZheLin]

@bryancsk Her name is Claude

[Alex @AlexZheLin]

The safest way to use LLMs is to get them to ask you questions, and even that isn't risk-free

[Alex @AlexZheLin]

@NickADobos Yeah although the "agentic" part scared me a little haha

[Alex @AlexZheLin]

@yacineMTB Easy, the ROI is infinity positive

[Alex @AlexZheLin]

@tszzl I dunno, there's also major disappointment with slop that makes me go back very naturally to raw dogging

[Alex @AlexZheLin]

@yacineMTB POV: Macro-ing in Starcraft X

[Alex @AlexZheLin]

@yacineMTB Same, just think how empty and pointless it would be otherwise

[Luca Dellanna @DellAnnaLuca]

I get where they’re coming from, but it’s like the laziest and most superficial solution.

The people who came up with this are supposed to be proficient with coding and care about education, and this is the best they can come up with?

The better version: force your students to use *your* LLM, which you specifically instructed to only be of support and not do the work. Have it log the prompts. Have it ask questions and log the answers. Evaluate the students on such logs, on top of the homework deliverable.

[Daniel Jeffries @Dan_Jeffries1]

The Infinite Stack.

Problems are endless. You abstract a lower level of problems and create more challenges and more complexity higher up the stack.

Complexity breeds complexity. You get more jobs and more varied jobs.

This goes on forever. It's endless.

[Alex @AlexZheLin]

@socoloffalex @uncertainsys Why not acqui-hire the builders of the apps? Integration is a bitch and the builders could pursue their vision.

Just seems like a stupid strat overall

[Alex @AlexZheLin]

@LinusEkenstam Isn't the other option that execs just enact the plans but keep people in the dark?

Seems better that they communicate their strategic plans with the staff? It's what I would prefer anyways.

[Alex @AlexZheLin]

@QiaochuYuan Stupid question but how is this even conceptually possible.

Like I get the "up and to the right" part but fundamentally it can't improve without lived experience with reality... right?

Isn't model collapse a thing?

[Alex @AlexZheLin]

Insight: Facts can't be vague but questions can.
Therefore (for complex tasks) the best and safest use of AI is to ask the user questions, not provide answers.

[Alex @AlexZheLin]

I think a big part of why people use AI as a thinking tool is because humans are just better at thinking through dialogue.

The problem is that current state AI is a dangerously overconfident dialogue partner.

[Alex @AlexZheLin]

D'un côté ça me semble très logique comme approche mais d'un autre côté je me questionne à savoir si l'IA ne pourrait pas aider dans mes réflexions.

À long-terme il semble évident que cela cause une érosion des capacités mais sur le "prochain problème", c'est comme s'empêcher d'aller chercher de l'aide qui pourrait être utile.

Je pense que l'application de cette règle deviendra aussi un défi en discipline... mais peut être bien nécessaire.