Leynos

10 ms·

181.
by Leynos 1y ago
At least for me, I can have five of these processes running at once. I can also use Deepresearch for generating the designs with a survey of literature. I can use NotebookLM to analyse the designs. And I use Sourcery, CodeRabbit, Codex and
182.
by Leynos 1y ago
That's what GPT-5 Pro and Grok 4 Heavy do. Those are the ones you pay triple digit USD a month for.
183.
by Leynos 1y ago
They don't see runs of spaces very well, so most of them are terrible at ASCII art. (They'll often regurgitate something from their training data rather than try themselves.) And unless their terminal details are included in the c
184.
by Leynos 1y ago
Codex web has a fun one where if you post multiple @codex comments to a PR, it gets confused as to which one it should be following because it gets the whole PR + comments as a homogenized mush in its context. I ended up rigging a userscri
185.
by Leynos 1y ago
They're counting model training as a capital expenditure, and so excluding it from running costs.
186.
by Leynos 1y ago
GPT-5 is a dab hand at Rust
187.
by Leynos 1y ago
So the author is complaining about people using copilot, not about copilot submitting comments and PRs automatically without the repo owner's consent?
188.
by Leynos 1y ago
I suspect it's either fantasy or fabrication.
189.
by Leynos 1y ago
The way BlueSky is structured, you have to follow people to see their content in your feed. Hence my confusion. Other people have pointed out that the "Discover" feed with a fresh account may show material that the OP finds unappe
190.
by Leynos 1y ago
Look at NovelCrafter. They have a great Discord community where people share their prompts and workflows. There is also the WritingWithAI subreddit.
191.
by Leynos 1y ago
That makes sense. My discover feed is pretty much all Scottish politics and tech.
192.
by Leynos 1y ago
Normally, I need to follow someone on BlueSky to see their content or content shared by them. It doesn't inject recommendations into my feed.
193.
by Leynos 1y ago
So how did the content appear in your feed?
194.
by Leynos 1y ago
DeepSeek is bad for hallucinations in my experience. I wouldn't trust its output for anything serious without heavy grounding. It's great for fantastical fiction though. It also excels at giving characters "agency".
195.
by Leynos 1y ago
Is it unreasonable not to want to share a social network with the people who want to exclude you from practical participation in society?
196.
by Leynos 1y ago
Search for "<topic of interest> starter pack"
197.
by Leynos 1y ago
Regular appearance as in posted by the people you follow?
198.
by Leynos 1y ago
That's interesting. I am curious about the extent of the training data in these models. I asked Kimi K2 for an account of growing up in my home town in Scotland, and it was ridiculously accurate. I then asked it to do the same for a si
199.
by Leynos 1y ago
Where is caching breaking so badly that this is happening? Are OpenAI failing to use etags or honour cache validity?
200.
Blink Twice: Measuring Strategic Deception Amongst LLMs(coup.khoj.dev)
2 points by Leynos 1y ago|0 comments
201.
by Leynos 1y ago
Mate, you've had since 1995 to come up with a functioning micropayments system, and you're unhappy that I am not requesting every image referenced by your web page? I think that we are long past the point where we can safely say t
202.
by Leynos 1y ago
I don't think chat context and memory count as "hidden features"
203.
by Leynos 1y ago
I'm a big proponent of AI as a tool for work, but unless you have a perfect received pronunciation accent, voice chat is painful. It's as if the AI chatbots were trained on Radio 4 and not much else.
204.
by Leynos 1y ago
Karpathy's definition definitely requires: 1. Not looking at the code 2. YOLO everything 3. Paste errors back into the model verbatim That said, I describe what I do as vibe coding, but I introduce code review bots into the mix. I als
205.
by Leynos 1y ago
As you say, your experience isn't universal, and we all have different modes of learning that work best for us.
206.
by Leynos 1y ago
I mostly use it because it makes rebasing a branch painless compared with CLI or the GitHub web interface. I was also using it at work when I managed a team because it let me see how their branches sat in relation to `main`.
207.
by Leynos 1y ago
MCP can be used as a form of context augmentation (i.e., RAG). It allows models to specify how that context augmentation is generated through tool use. It's a formalized way of allowing developers to implement tools (using JSON-RPC) i
208.
by Leynos 1y ago
Obsidian Google (for Deep Research, Gemini and NotebookLM) ChatGPT (For ChatGPT and Codex) GitKraken CodeRabbit Novelcrafter (for fiction) Wondercraft (wrapper around Elevnlabs and Googles TTS) Seriously considering Elevenreader as well Rai
209.
by Leynos 1y ago
I was happy seeing that someone else is taking this approach. I throw Sourcery and Codescene into the mix as well. Sourcery sees things from a more architectural stand point, while Codescene is great for catching code quality deterioration
210.
How to Supercharge "Vibe Coding" with CodeRabbit and OpenAI Codex(medium.com)
2 points by Leynos 1y ago|1 comments

More ›