Ryan Mcgarvey
@ryan.mcgarvey.social
He/Him
Engineer @ Storyboard.com
Side project: roletogether.com
Serious goose
Engineer @ Storyboard.com
Side project: roletogether.com
Serious goose
"Using AI" is a problematically ambiguous phrase. Smart autocomplete is technically AI. Semantic search of a code base is AI.
87% sounds reasonable as there's always some fraction of SWEs that are obstinate about anything AI
87% sounds reasonable as there's always some fraction of SWEs that are obstinate about anything AI
August 18, 2025 at 6:31 PM
"Using AI" is a problematically ambiguous phrase. Smart autocomplete is technically AI. Semantic search of a code base is AI.
87% sounds reasonable as there's always some fraction of SWEs that are obstinate about anything AI
87% sounds reasonable as there's always some fraction of SWEs that are obstinate about anything AI
Not to mention that the average person doesn't put a lot of effort validating what they read on the internet.
ChatGPT only needs to be less wrong than the average person's ability to search to have a net positive effect. It's a low bar.
ChatGPT only needs to be less wrong than the average person's ability to search to have a net positive effect. It's a low bar.
May 19, 2025 at 3:52 AM
Not to mention that the average person doesn't put a lot of effort validating what they read on the internet.
ChatGPT only needs to be less wrong than the average person's ability to search to have a net positive effect. It's a low bar.
ChatGPT only needs to be less wrong than the average person's ability to search to have a net positive effect. It's a low bar.
Perhaps I misunderstood how electron works. The docs all say ~ updating the app after it is built is not officially supported.
So if you're inside a strict IT environment where someone needs to validate each signed installer, it can be cumbersome to use the official updater.
So if you're inside a strict IT environment where someone needs to validate each signed installer, it can be cumbersome to use the official updater.
May 1, 2025 at 1:10 PM
Perhaps I misunderstood how electron works. The docs all say ~ updating the app after it is built is not officially supported.
So if you're inside a strict IT environment where someone needs to validate each signed installer, it can be cumbersome to use the official updater.
So if you're inside a strict IT environment where someone needs to validate each signed installer, it can be cumbersome to use the official updater.
Amazing!
We are in the process of shopping for electron like solutions.
Are you planning to officially support local updates? I.e. not require a full re-install of the app, but it just updates the internal app and reboots.
We are in the process of shopping for electron like solutions.
Are you planning to officially support local updates? I.e. not require a full re-install of the app, but it just updates the internal app and reboots.
May 1, 2025 at 12:13 PM
Amazing!
We are in the process of shopping for electron like solutions.
Are you planning to officially support local updates? I.e. not require a full re-install of the app, but it just updates the internal app and reboots.
We are in the process of shopping for electron like solutions.
Are you planning to officially support local updates? I.e. not require a full re-install of the app, but it just updates the internal app and reboots.
Hit and miss, but not nothing.
May 1, 2025 at 3:35 AM
Hit and miss, but not nothing.
Broadly speaking on the product side, anytime we see a problem that used to be a dead end because the data isn't structured well enough or is otherwise too ambiguous, it's an opportunity for AI, especially if the outcome can tolerate some degree of failure.
Content moderation is a good example.
Content moderation is a good example.
April 7, 2025 at 7:18 PM
Broadly speaking on the product side, anytime we see a problem that used to be a dead end because the data isn't structured well enough or is otherwise too ambiguous, it's an opportunity for AI, especially if the outcome can tolerate some degree of failure.
Content moderation is a good example.
Content moderation is a good example.
We also use it to refactor large sections of code. Anything that's rote, mechanical work that might have taken an engineer a few days to type out and therefore never happens, can get done inside an hour with the right AI flow. It changes the cost benefit dynamics of certain work.
April 7, 2025 at 7:14 PM
We also use it to refactor large sections of code. Anything that's rote, mechanical work that might have taken an engineer a few days to type out and therefore never happens, can get done inside an hour with the right AI flow. It changes the cost benefit dynamics of certain work.
In addition to the usual translation, Text to speech, speed to text:
Fielding questions with natural language search - a very difficult problem that with the right prompt and context gets you very accurate results.
Or large scale categorization of squishy data. Like sentiment analysis type stuff.
Fielding questions with natural language search - a very difficult problem that with the right prompt and context gets you very accurate results.
Or large scale categorization of squishy data. Like sentiment analysis type stuff.
April 7, 2025 at 7:10 PM
In addition to the usual translation, Text to speech, speed to text:
Fielding questions with natural language search - a very difficult problem that with the right prompt and context gets you very accurate results.
Or large scale categorization of squishy data. Like sentiment analysis type stuff.
Fielding questions with natural language search - a very difficult problem that with the right prompt and context gets you very accurate results.
Or large scale categorization of squishy data. Like sentiment analysis type stuff.
Yeah, somehow "being an engineer helps you do the thing" got confused with "doing the thing makes you an engineer".
April 1, 2025 at 11:52 AM
Yeah, somehow "being an engineer helps you do the thing" got confused with "doing the thing makes you an engineer".
That's a big assumption that you're reading into the fluctuations. We have now idea if they are reacting to AI investment or something else. Could be that they are better positioned to survive a number of horrible potential things coming down the pike.
March 26, 2025 at 10:26 PM
That's a big assumption that you're reading into the fluctuations. We have now idea if they are reacting to AI investment or something else. Could be that they are better positioned to survive a number of horrible potential things coming down the pike.
That's not really true tho? Those lease cancelations aren't really an indication of slowing down so much as how the process works for expanding.
They are spending $80B this year on data centers. They're still betting the farm on AI. Everyone is.
They are spending $80B this year on data centers. They're still betting the farm on AI. Everyone is.
March 26, 2025 at 9:56 PM
That's not really true tho? Those lease cancelations aren't really an indication of slowing down so much as how the process works for expanding.
They are spending $80B this year on data centers. They're still betting the farm on AI. Everyone is.
They are spending $80B this year on data centers. They're still betting the farm on AI. Everyone is.
Yes. Being people over.
March 25, 2025 at 10:39 AM
Yes. Being people over.
If the musk administration is canceling ngad in favor of short range drones, which is nonsense when you care about facing off against china in the pacific... maybe that's the point. They're not preparing for a war in the pacific but to use them on the local population.
They.mr
February 9, 2025 at 10:55 PM
If the musk administration is canceling ngad in favor of short range drones, which is nonsense when you care about facing off against china in the pacific... maybe that's the point. They're not preparing for a war in the pacific but to use them on the local population.
Did it actually train for $5M?
I thought it *fine tuned* for that much.
Meta paid for the training.
I thought it *fine tuned* for that much.
Meta paid for the training.
January 28, 2025 at 5:19 PM
Did it actually train for $5M?
I thought it *fine tuned* for that much.
Meta paid for the training.
I thought it *fine tuned* for that much.
Meta paid for the training.
I'll note, it's nearly perfect, but not quite.
For instance, using a require exactly one from a set of inputs, and then passing those inputs to another function that has require exactly one fails.
For instance, using a require exactly one from a set of inputs, and then passing those inputs to another function that has require exactly one fails.
January 15, 2025 at 12:47 PM
I'll note, it's nearly perfect, but not quite.
For instance, using a require exactly one from a set of inputs, and then passing those inputs to another function that has require exactly one fails.
For instance, using a require exactly one from a set of inputs, and then passing those inputs to another function that has require exactly one fails.
Is that the worst outcome? Canceling credit for people who probably shouldn't have it anyway, if the only way for them to have it is with crazy high rates?
Not saying I agree with the proposal, but it sounds similar to if you can't afford to pay minimum wage you can't afford to be in business.
Not saying I agree with the proposal, but it sounds similar to if you can't afford to pay minimum wage you can't afford to be in business.
December 31, 2024 at 1:09 AM
Is that the worst outcome? Canceling credit for people who probably shouldn't have it anyway, if the only way for them to have it is with crazy high rates?
Not saying I agree with the proposal, but it sounds similar to if you can't afford to pay minimum wage you can't afford to be in business.
Not saying I agree with the proposal, but it sounds similar to if you can't afford to pay minimum wage you can't afford to be in business.
With what money? Under whose direction?
The investment is being made by people that think these aren't parlor tricks.
The investment is being made by people that think these aren't parlor tricks.
December 26, 2024 at 1:28 PM
With what money? Under whose direction?
The investment is being made by people that think these aren't parlor tricks.
The investment is being made by people that think these aren't parlor tricks.
Isn't the relative power consumption quite low?
What if these parlor tricks come with a shit tin of green energy investment that will remain long after the trick is finished?
What if these parlor tricks come with a shit tin of green energy investment that will remain long after the trick is finished?
December 26, 2024 at 1:17 PM
Isn't the relative power consumption quite low?
What if these parlor tricks come with a shit tin of green energy investment that will remain long after the trick is finished?
What if these parlor tricks come with a shit tin of green energy investment that will remain long after the trick is finished?
Oh right - so he's not saying they have a specific crash site, just that there's a program outside of congressional over site related to UAP and he doesn't like it. Still interesting tho.
December 22, 2024 at 4:33 PM
Oh right - so he's not saying they have a specific crash site, just that there's a program outside of congressional over site related to UAP and he doesn't like it. Still interesting tho.
Schumer does? Curious do you have a quote / source for him saying he believes this? That absolutely would change my stance on this a bit.
December 22, 2024 at 4:10 PM
Schumer does? Curious do you have a quote / source for him saying he believes this? That absolutely would change my stance on this a bit.
Wasn't Contrarian Dullard the very *first* job to be replaced with AI? What with it requiring the lowest form of intelligence to replicate.
December 20, 2024 at 2:13 PM
Wasn't Contrarian Dullard the very *first* job to be replaced with AI? What with it requiring the lowest form of intelligence to replicate.