bengarney

5K posts

bengarney banner
bengarney

bengarney

@bengarney

@marbleitup developer. 🔮 I run https://t.co/1VYoeKXAU7 - we solve hard problems for you. Past projects: Loom SDK, Pushbutton Labs, Blockland, Garage Games.

eugene, or, usa Katılım Eylül 2007
805 Takip Edilen2K Takipçiler
Sabitlenmiş Tweet
bengarney
bengarney@bengarney·
What makes old franchises live or die? I wrote about my personal experiences with Tribes, Marble Blast/Marble It Up, Blockland, and Zap - all games with 20+ year followings. bengarney.com/2025/05/15/seq…
English
2
2
9
497
bengarney
bengarney@bengarney·
@wackie The best and highest use of the features of the browser is to implement a manageable subset of the features of the browser much better than the browser.
English
1
0
1
16
bengarney
bengarney@bengarney·
@wackie Tbh me too though for me it was vb4
English
0
0
0
12
JD Conley
JD Conley@wackie·
i miss yolo releasing zipped up vb3 code to aol chatrooms
English
1
0
2
46
bengarney
bengarney@bengarney·
More gems.
bengarney tweet mediabengarney tweet mediabengarney tweet media
English
0
0
2
73
bengarney
bengarney@bengarney·
I needed to generate test data for a real estate project, and ended up with some delightfully unhinged listings.
bengarney tweet mediabengarney tweet mediabengarney tweet mediabengarney tweet media
English
1
0
2
91
bengarney
bengarney@bengarney·
@wackie First you doubt the AI. Then you use the AI. Finally, you ignore the AI’s output completely.
English
0
0
0
18
JD Conley
JD Conley@wackie·
who else uses their voice ai for rubber ducking? i start dictating a prompt in cursor only to solve my own problem and then just go fix it.
JD Conley tweet media
English
1
0
0
76
JD Conley
JD Conley@wackie·
my followers can’t see my @OpenAI sora videos on my profile or feed in the app. i’m all alone 😭
English
1
0
0
33
bengarney
bengarney@bengarney·
@wackie If you’re in the training set it really does well. Flattened training code seems to work.
English
0
0
0
10
JD Conley
JD Conley@wackie·
@bengarney have been pleasantly surprised. so nice to just have the code sitting there. i was like 'implement me a streaming base64 decoder so i can efficiently transform/stream this giant image blob i got in json from the ai using less memory' and, yah know, it just did it.
English
1
0
0
17
JD Conley
JD Conley@wackie·
i’m just going to raw dog hono server from now on. have ai write anything i need. team #nodependencies
English
1
0
0
60
JD Conley
JD Conley@wackie·
thanks Mac OS Tahoe, YES! i did want to use my one time code in my @cursor_ai agent prompt and when i came here to post about it.
JD Conley tweet mediaJD Conley tweet media
English
2
0
0
149
bengarney
bengarney@bengarney·
@wackie Big difference between that offer for 1yr experience and 20yr, that’s for sure. My first job wasn’t massively far from that - a little less and a little later stage - I didn’t think twice about the salary and lived in my boss’ garage for a long time… it worked out very well.
English
0
0
2
23
JD Conley
JD Conley@wackie·
my first programming job was for $9/hour after i dropped out of junior college – roughly double the minimum wage at the time. things worked out just fine. as long as they're taking a risk on the inexperienced and hungry i don't see an issue with it. have to get your foot in the door in the industry somehow.
Steve the Beaver@beaversteever

average YC startup offer

English
1
0
0
135
bengarney
bengarney@bengarney·
Fascinating to see multiple LLMs converging on the same answer for a fairly complex question. Although notice that it's only specifying the mantissa not the exponent...
Dave W Plummer@davepl1968

If you were writing the simulation we might live in, how many bits would you need in your data types? I asked @Grok and @ChatGPT both the following question. Answers in thread! "How many bits of precision would a floating point type need to have to represent the position of an electron in the observable universe down to a precision of the Planck distance?"

English
0
0
1
170
bengarney
bengarney@bengarney·
@wackie Obv you should be asking openai to generate refreshing nature art so you can be recharged without stepping away from managing your 12 Claude agents
English
0
0
0
24
zack
zack@zack_overflow·
I find it mind-blowing that everyday I work on a ~600k lines codebase It's impossible to comprehend how that amount of code even got written in the first place
zack tweet media
English
4
0
100
9.2K
bengarney
bengarney@bengarney·
@shapesandlines This is the true Oregon coast experience. Accept no substitutes. 🌦️
English
0
0
0
30
todd pickens
todd pickens@shapesandlines·
I'm headed to the cost for a couple of days, where it's expected to be...checks forecast... 🤷‍♂️
todd pickens tweet media
English
2
0
5
233
bengarney
bengarney@bengarney·
LLMs are getting better every month, so I still think there could be a breakthrough which changes everything. But the odds are gradually decreasing - the curve on the current generation is visibly leveling out.
English
0
0
0
102
bengarney
bengarney@bengarney·
Each step up that chain is a big win, but none have obviated the need for good teams, or arguably even reduced team size. Will agentic coding achieve a 10x reduction in staff? MSFT employee count has grown every year for the past decade.
English
1
0
0
117
bengarney
bengarney@bengarney·
"[LLMs] will only be truly impressive in the hands of exceptional software engineers" - if they require an incredibly talented user, how are LLMs different than a high friction text editor? That same user could use toggle switches... punch cards... assembly... (cont.)
English
1
0
1
130
bengarney
bengarney@bengarney·
To expand, there will be a wave of deliveries from nvidia to data centers that will not have the demand to use them. Frontline use will be on local devices, and data centers will be for bulk usage or very high end models.
English
0
0
0
93
bengarney
bengarney@bengarney·
The trend in GenAI over the next few years will be less with more. Apple has a big incentive to make their deployed hardware more effective, thus research like arxiv.org/abs/2505.19640… There will be a lot of underutilized AI data centers because of this.
English
1
0
3
153