Kyle Hart

engineer, entrepreneur, digital health developer (dx/rx/tx), genome scientist, data wrangler

[Kyle Hart @kylehart]

You wouldn't dump raw google search results into a chat session with your friend - why would you do the same with LLM answers? It's impolite! Don't be a Meat Puppet. Curate prompts, curate answers! Do the first super-well and possibly skip the second!

[Peter Steinberger 🦞 @steipete]

I’ll totally use this in my next presentation.

[Jeff Dean @JeffDean]

Announcing Discovery Loop!

I am very excited to announce that, along with my longtime friends and collaborators @Sanjay_Ghemawat, @OriolVinyalsML and @quocleix, we are founding Discovery Loop (@DiscoLoopAI), a Public Benefit Corporation whose mission is to automate machine learning, science, and engineering to accelerate discoveries and progress. The four of us have worked together for 14 to 30 years, and have helped build some of the world’s most used products, infrastructure and AI models, and we’re excited to turn our attention to this ambitious endeavor.

♾

Learn more at: https://t.co/Rv3LMdLluK

[Kyle Hart @kylehart]

@thedankoe that's a banger, but i will go with - the only source of _discovered_ truth is making mistakes. lookup is not discovering. experiments reveal findings. or failure is the quintessential form of truth.

[Kyle Hart @kylehart]

@gokulr i think it's tolerable if you Fire and Return. increasingly this is the ideal flow one strives for?

[Joanna Stern @JoannaStern]

In a review of my household safety evaluations, I identified five incidents in which my child escaped the sandbox, reached the kitchen and gained unauthorized access to the snacks. The incidents occurred 16 months ago but has only now come to my attention.

This post explains what happened, how it happened and why my child is better than your child at everything.

[Kyle Hart @kylehart]

You may not be a scientist or an engineer, but you have been warned about power tools. They will eventually harm you if used wrongly in a repeated way.

Pay attention and add intention! Be self-aware and self-critical. Verify the outcomes to build trust. Only YOU can do this.

[Kyle Hart @kylehart]

Getting answers from the Oracle at Delphi or the Swami on the mountain top required substantial effort - you had to suffer a long journey and sharpen your prepared question as you traveled.

So it is with modern AI.

You are doing it wrong when you expect easy answers. /1

[Kyle Hart @kylehart]

The Oracle fools you with beautiful prose and confidence, from meager prompts, starved for clarity, intentions, and details.

We complain about inaccuracy while abdicating responsibility for essential inputs. For even a rudimentary understanding of how the Oracle works.

[Kyle Hart @kylehart]

Two words: Contagious Cancer

Yikes! This is very bad.
Also: New Alien Invaders from Canada
[literally true!]

https://t.co/gFxOjfAjjA

[Kyle Hart @kylehart]

@willwashburn iphone auto-transcription of voicemail saved me from never leaving voicemails (for this reason). good use of technology - liberates the talkers to be themselves, oblivious to what their audience might prefer.

[Kyle Hart @kylehart]

We don't spellcheck LLM instructions because it's neither impolite nor necessary, but that leads us toward having two "speech" personas.

Will we adapt to the extra cognitive load of switching? Lean in, and fall into semi-coherent rambling speech if it still works?

#botilect

[Josh Gray @joshfgray_]

the future of healthcare is patient to agent

[Kyle Hart @kylehart]

@aakashgupta hibernating bears have the same amonia waste treatment cycle.

[Kyle Hart @kylehart]

@joshuapliu Regulatory-adjacent markets (DTC) are a lightning rod for (exaggerated) paternal "harm" criticisms. Huge population health benefits from this tech, but no pop health champions here! DTC is provably good health policy even including the (possible) mental anguish of False Pos.

[Kyle Hart @kylehart]

Calling out someone for their AI "slop" in another way of saying:
= "I don't want to learn AI" - so you shouldn't use it either
= "I don't want to read anything that doesn't have human errors in it" - it makes me uncomfortable when people correct their formerly flawed writing

[Delip Rao e/σ @deliprao]

In-context learning in LLMs

[Kyle Hart @kylehart]

@Shpigford Is this pre-AI thinking? Switching costs of natural language applications are small to nil. These are not your father's API lock-ins. You have to re-skill constantly even if you DO stick with the same vendor, so is that much easier than switching vendors? Not a lot.

[Kyle Hart @kylehart]

Codex starts answering, realizes it can do better, the rewrites the response!

Kudos to the Product manager who said "Leave that in - let them see it" or even better expressly chose it.

We are the Watchers
Not the Wizards
Teach us Your Magic
So we may harness it and guide you.

[Jonathan Ross @JonathanRoss321]

My job used to be writing code. Now it's marriage counseling for Codex and Claude Code.

[Craig Fuller 🛩🚛🚂⚓️ @FreightAlley]

Yesterday, I spoke with the CEO of a mega fleet, who said most of his truckload business was doing well, except for one segment: food & beverage.

He called the lack of volume from this segment "unusual."

I told him we believed GLP-1s were causing a significant slowdown in food and beverage shipments, as we had just completed a market study on GLP-1's impact on freight shipments.

Our study, now published in a SONAR Sitrep, available online, estimates that 851k truckloads have been removed from the market due to GLP-1s, and this number could ramp to 1.95m by 2030.

Not only are Americans getting skinnier. Their truckloads are as well.