Max Mohr @MaxCreatesCode

Coding a better world. Software Engineer (HPC) Georgia Tech, UNCC, USC Joined April 2010

Tweets

170
Followers

52
Following

263
Likes

9K

Max Mohr @MaxCreatesCode

23 hours ago

@thsottiaux “Workplace routing discovery timed out”, makes a ton of personal tools built around codex useless.

0 0 1 184 0

View Details

Max Mohr @MaxCreatesCode

a day ago

@cheatyyyy @abliteration_ai The abliterated models are free on hugging face? Even if you needed to rent hardware to host the model it’s like $12/hr …what money do you think you need to spend???

0 0 0 176 0

View Details

@LLMJunky @JesusIsGod Much larger unified memory systems aiming to come out within the next two years, you will always be able to fit a much more capable model into 1-2TB of DDR5X compared to a DGX Spark, GH200NVL2 …there just isn’t a good product at $40k yet …512GB Mac Studio & 192GB AMD Halo

0 0 0 59 0

View Details

Max Mohr @MaxCreatesCode

2 days ago

@theo OmniRoute's consensus mode does this really well as well, could probably trade notes.

0 0 1 58 0

View Details

Max Mohr @MaxCreatesCode

2 days ago

@jonnym1ller Not a cyber weapon, only dangerous to companies that pretend security through obscurity is real. Gatekeeping misaligned knowledge is far more dangerous in this context. Pretty easy to write code that is “actually secure” unlike places that offload legal liability to red teams.

0 0 1 5 0

View Details

Max Mohr @MaxCreatesCode

3 days ago

@MrRoflWaffles Run tracert when the latency spike is happening, the output will show them if it is their equipment at fault & they will trust it. Figure out where you are uploading to or gaming to send it too, you can use this command pointed at CoD servers for example.

0 0 0 122 0

View Details

Max Mohr @MaxCreatesCode

4 days ago

@kunchenguid @utpalsinghdev That would not be the same as a permanent cache at all. Storing the encoded form & latent space representations has much more meaning than the words alone can convey…work was done to build up that meaning. Won’t be compatible with future models, gets thrown out then.

0 0 1 69 0

View Details

Max Mohr @MaxCreatesCode

4 days ago

@Macjam0 @maria_rcks No, the cache is the encoded form of a conversation. Converting it back to words loses a lot of meaning every time they do it which is why it is not preserved at each step. They simply made a way to offload this form of the conversation to hard drives to load it back directly.

1 0 0 1K 0

View Details

Max Mohr @MaxCreatesCode

4 days ago

@thsottiaux Actually huge.

0 0 1 35 0

View Details

Max Mohr @MaxCreatesCode

6 days ago

@CampbellKaleb23 @thsottiaux 100% ragebait since we could already do this if you used it most other places… I guess they hide chain of thought now but this is a feature we used to have & on their side they store it encoded so it would be just truncating at those points. So frustrating.

0 0 6 2K 1

View Details

Max Mohr @MaxCreatesCode

a week ago

@akashjayn @thsottiaux As in we only got the grandfathered capacity for a single week. The reset did not give us the same capacity across any of my accounts, some of which are running repeated 6.1-Sol workloads.

0 0 1 16 0

View Details

Max Mohr @MaxCreatesCode

a week ago

@akashjayn @thsottiaux No shot he believes us but I hope they actually look into it because it is a very real change from my metrics on 6.1-Sol the week prior to now. Would be very obvious if they actually looked 🙏 issue with most "grandfathered" 20x accounts

1 0 0 15 0

View Details

Max Mohr @MaxCreatesCode

a week ago

@SourceCodeplz @NVIDIAAI Free UltraFast model is wild, way faster than Sol-Ultrafast which was introduced as a $500/month feature less than a month ago.

0 0 0 74 0

View Details

Max Mohr @MaxCreatesCode

a week ago

@akashjayn @thsottiaux Because the capacity of the 20x plan was meaningfully downgraded one week after he said 20x accounts would keep their grandfathered capacity for a month… the resets have covered this gap for most users but still weird to not keep their word, might be a bug.

1 0 1 11 0

View Details

Max Mohr @MaxCreatesCode

a week ago

@Angaisb_ gpt-image-2.5-sunburst supports custom resolutions …did you forget what Pixels are? No need for masking. “support custom dimensions as WIDTHxHEIGHT strings” developers.openai.com/api/docs/guide…

0 0 1 357 1

View Details

Max Mohr @MaxCreatesCode

a week ago

@VictorTaelin I thought you were raising? Could easily change the world in a few months with funding, that's not slow.

0 0 0 179 0

View Details

Max Mohr @MaxCreatesCode

a week ago

@bryan_johnson You are missing the mapping from genes to the purposes they actually serve (unfortunately they serve overlapping purposes) we have ways to edit genes inside living organisms & ways to fix the end caps on genes for people 120yo+ youtu.be/_cLLVQL0Kcs

0 0 0 148 0

View Details

Max Mohr @MaxCreatesCode

a week ago

@shauseth Succinctly conveying a ton of information in a form that is easy to comprehend is a sign of higher intelligence, many models & people are failing at this right now. Keeping it simple takes more effort, not less.