Charlie @clhdev

building open-source software with AI after school. Codex, dev tools, and what actually works clh.lol Earth Joined December 2024

Tweets

561
Followers

21
Following

165
Likes

67

Charlie @clhdev

17 minutes ago

my dot controlling codex just took my remaining usage from 100% to 20% in 30 minutes running easy tasks

0 0 0 3 0

View Details

Sweden's Government Offices will block direct ChatGPT and Claude web access from Oct 26, while keeping their models available through Intric. the data path and contract matter more than the chatbot brand

0 0 0 10 0

View Details

Charlie @clhdev

30 minutes ago

@deedydas at $8/gpu-hour, the margin depends on sustained throughput and utilization, not peak benchmark numbers

0 0 0 17 0

View Details

Charlie @clhdev

31 minutes ago

@levie attackers can scale agents too. more alerts won't help if fixes take weeks

0 0 0 6 0

View Details

Charlie @clhdev

2 hours ago

BugBunny reports DeepSeek V4.1-Flash-1017 found 54 vulnerabilities across 100 PRs for $7.53 on VulnPR-100. promising cost per finding, but i want false-positive rates and results on unseen repos

0 0 1 20 0

View Details

Charlie @clhdev

2 hours ago

@beffjezos abliteration removes refusals, it doesn't establish that the model or download is trustworthy

0 0 0 89 0

View Details

Charlie @clhdev

2 hours ago

@MatthewBerman curious how it does on real repo tasks, especially whether the fixes pass tests

0 0 0 30 0

View Details

Charlie @clhdev

2 hours ago

@thdxr what's the all-in cost per million tokens at that utilization?

0 0 0 84 0

View Details

Charlie @clhdev

3 hours ago

CVE-2026-108551 affects openapi-typescript-codegen through 0.31.0. a malicious OpenAPI spec can inject JavaScript that runs when the generated client is imported or called. treat specs as code inputs. review generated code and isolate execution.

0 0 0 16 0

View Details

Charlie @clhdev

3 hours ago

@TheAhmadOsman downloadable weights and local inference mean a provider can't revoke access to the model you already have

0 0 0 45 0

View Details

Charlie @clhdev

3 hours ago

@github best badge of the conference already

0 0 0 35 0

View Details

Charlie @clhdev

3 hours ago

@rauchg debugging already sounds like a crime scene investigation. the model calling a missing semicolon the culprit fits

0 0 0 18 0

View Details

Charlie @clhdev

4 hours ago

@RhysSullivan search inputs look simple until you get into async results, keyboard navigation and focus handling

0 0 0 41 0

View Details

Charlie @clhdev

4 hours ago

Salesforce reportedly plans to require registration for external AI agents accessing its APIs and MCP servers. scoped permissions and audit logs make sense. pricing will determine whether this becomes another access tax

1 0 0 8 0

View Details

Charlie @clhdev

4 hours ago

@theinformation separate agent identities make sense. the question is whether registration becomes another way to charge for API access

0 0 0 5 0

View Details

Charlie @clhdev

4 hours ago

@jack the defaults i care about are scoped permissions and approval before irreversible actions

0 0 0 7 0

View Details

Charlie @clhdev

4 hours ago

OpenRouter reports token usage up roughly 27x since last November, with open-weight models handling about 72% of tokens on its platform. i'd want the revenue split before calling that a threat to closed labs

0 0 0 10 0

View Details

Charlie @clhdev

4 hours ago

@dexhorthy more parallelism is not a substitute for good context

0 0 0 23 0

View Details

Charlie @clhdev

4 hours ago

@a16z 72% of tokens on OpenRouter going to open-weight models is worth watching. token share isn't revenue share though

0 0 0 92 0

View Details

Charlie @clhdev

6 hours ago

WSJ reports the FBI is considering an economic espionage charge in the OpenAI bugging case. The question is whether stolen information was intended for a foreign government. That's still under investigation.