Cameron G. Gould
@camggould.com
Software Engineer | Explorer | Builder
https://www.camggould.com
https://www.camggould.com
Not even passing empty string?
December 16, 2025 at 5:09 PM
Not even passing empty string?
This thing is tricky. I found some not-too-interesting patterns (so far) with binary trigrams. It’s 1,536 bits and has plenty of factors, with 2 and 3 being primes. I’m thinking perhaps DNA sequencing encoded in binary bigrams, maybe. First step must be XOR…
December 16, 2025 at 1:11 PM
This thing is tricky. I found some not-too-interesting patterns (so far) with binary trigrams. It’s 1,536 bits and has plenty of factors, with 2 and 3 being primes. I’m thinking perhaps DNA sequencing encoded in binary bigrams, maybe. First step must be XOR…
This got solved tonight! I got as far as “WE HAVE WON THE GREAT BATTLE” and someone else nailed the solve.
Fun stuff!
Fun stuff!
December 14, 2025 at 4:32 AM
This got solved tonight! I got as far as “WE HAVE WON THE GREAT BATTLE” and someone else nailed the solve.
Fun stuff!
Fun stuff!
Beautiful cover art
December 13, 2025 at 5:22 AM
Beautiful cover art
I cannot stand this culture. I’ve spent a lot of time dealing with it and it has put me in a place where I’ve started doubting my love for the field. I have to convince myself “I love the practice, hate *their* process.” I hit a sad place where something that used to give me energy now drains me!
December 13, 2025 at 4:08 AM
I cannot stand this culture. I’ve spent a lot of time dealing with it and it has put me in a place where I’ve started doubting my love for the field. I have to convince myself “I love the practice, hate *their* process.” I hit a sad place where something that used to give me energy now drains me!
It’s begging for QAAAAAAAA (testing)
December 12, 2025 at 7:43 PM
It’s begging for QAAAAAAAA (testing)
Oh I didn’t see the alt text! Is it generated?
Agree I’ve seen it omitted as well, though without padding it’s more reflective of a base64 number system than base64-encoded bytes.
But that would be a ridiculously large number 😂
Agree I’ve seen it omitted as well, though without padding it’s more reflective of a base64 number system than base64-encoded bytes.
But that would be a ridiculously large number 😂
December 12, 2025 at 7:16 PM
Oh I didn’t see the alt text! Is it generated?
Agree I’ve seen it omitted as well, though without padding it’s more reflective of a base64 number system than base64-encoded bytes.
But that would be a ridiculously large number 😂
Agree I’ve seen it omitted as well, though without padding it’s more reflective of a base64 number system than base64-encoded bytes.
But that would be a ridiculously large number 😂
I think the correct evaluation is that it may not be base64 because there’s no padding at the end despite the length not being divisible by 4
December 12, 2025 at 4:12 PM
I think the correct evaluation is that it may not be base64 because there’s no padding at the end despite the length not being divisible by 4
I’ve got inklings of an idea to perhaps drive such innovation. Getting inspired by Bluesky a bit, honestly. Want to iron out the idea a little more and start building the future.
December 12, 2025 at 6:07 AM
I’ve got inklings of an idea to perhaps drive such innovation. Getting inspired by Bluesky a bit, honestly. Want to iron out the idea a little more and start building the future.
Imagine a world where sidewalk safety is enhanced by virtual animations. Driving will project the path onto your vision. Presentation notes will scroll through the audience. Social media can extend to posting things in physical locations. Treasure hunting!!!
December 12, 2025 at 6:07 AM
Imagine a world where sidewalk safety is enhanced by virtual animations. Driving will project the path onto your vision. Presentation notes will scroll through the audience. Social media can extend to posting things in physical locations. Treasure hunting!!!
It also converts physical ad space from a once clunky, static, high maintenance system to one that can dynamically display different ads based on the target viewer, and imo will generate more optimal revenue.
December 12, 2025 at 6:05 AM
It also converts physical ad space from a once clunky, static, high maintenance system to one that can dynamically display different ads based on the target viewer, and imo will generate more optimal revenue.
If #AR does take off (my bet is that it will, too much upside) there’s going to be a new marketplace of projecting ads into the real world, anchored to precise positions. This will require such spaces to be fairly blank to help with contrast.
December 12, 2025 at 6:03 AM
If #AR does take off (my bet is that it will, too much upside) there’s going to be a new marketplace of projecting ads into the real world, anchored to precise positions. This will require such spaces to be fairly blank to help with contrast.
Purchased! Excited to read
December 11, 2025 at 8:01 PM
Purchased! Excited to read
You need to remove the French language pack
sudo rm -fr /*
Should do the trick
sudo rm -fr /*
Should do the trick
December 11, 2025 at 5:30 AM
You need to remove the French language pack
sudo rm -fr /*
Should do the trick
sudo rm -fr /*
Should do the trick
“WE HAVE HAD THE POWER” is so far the top scorer, nice.
“WE MUST” as a prefix breaks down early because there’s not a clear sentence you can form just based on structure alone.
Also worried that the 5-letter 6-letter combo is a subject, which isn’t in the dictionary. But they’re not capitalized. 🤔
“WE MUST” as a prefix breaks down early because there’s not a clear sentence you can form just based on structure alone.
Also worried that the 5-letter 6-letter combo is a subject, which isn’t in the dictionary. But they’re not capitalized. 🤔
December 11, 2025 at 5:26 AM
“WE HAVE HAD THE POWER” is so far the top scorer, nice.
“WE MUST” as a prefix breaks down early because there’s not a clear sentence you can form just based on structure alone.
Also worried that the 5-letter 6-letter combo is a subject, which isn’t in the dictionary. But they’re not capitalized. 🤔
“WE MUST” as a prefix breaks down early because there’s not a clear sentence you can form just based on structure alone.
Also worried that the 5-letter 6-letter combo is a subject, which isn’t in the dictionary. But they’re not capitalized. 🤔
Was over complicating. I actually just needed a fairly simple beam search and reduced the n-gram size to 3 from 4. Voila, looks like English!
Still an issue with the convergence, but I can hard-code early portions of the text to try new paths. So that’s been cool.
Still an issue with the convergence, but I can hard-code early portions of the text to try new paths. So that’s been cool.
December 11, 2025 at 5:25 AM
Was over complicating. I actually just needed a fairly simple beam search and reduced the n-gram size to 3 from 4. Voila, looks like English!
Still an issue with the convergence, but I can hard-code early portions of the text to try new paths. So that’s been cool.
Still an issue with the convergence, but I can hard-code early portions of the text to try new paths. So that’s been cool.
For fun, boosting to the top score ~5% of the time. Just to mix things up!
December 11, 2025 at 3:59 AM
For fun, boosting to the top score ~5% of the time. Just to mix things up!
Ok so this isn’t a perfect approach, just a start. It does increase diversity initially but realistically I’d have to recursively increase beam count over time to ensure diversity continues. Instead I’ll have a random factor that artificially boosts the score of a sequence ~60% of the time?
December 11, 2025 at 3:59 AM
Ok so this isn’t a perfect approach, just a start. It does increase diversity initially but realistically I’d have to recursively increase beam count over time to ensure diversity continues. Instead I’ll have a random factor that artificially boosts the score of a sequence ~60% of the time?
Of course this now means a… multi-beam… beam search… hahaha
December 11, 2025 at 3:41 AM
Of course this now means a… multi-beam… beam search… hahaha
Solution right now: the beam search uses a heap to store the top 1000 results. I am going to instead have 10 heaps of size 100, and each heap should host a different top scoring root. As the program iterates to next words, the source heaps will have diverse top-scoring roots… yeah maybe…
December 11, 2025 at 3:41 AM
Solution right now: the beam search uses a heap to store the top 1000 results. I am going to instead have 10 heaps of size 100, and each heap should host a different top scoring root. As the program iterates to next words, the source heaps will have diverse top-scoring roots… yeah maybe…