81–90 of 96 posts
Earlier quoted context omitted.
> Name one company selling AI models who didn't steal their data or aquire it through dubious or unethical means? That's what I said. I was absolutely referring to OpenAI and Anthropic as well as the Chinese models.
I misunderstood you. I am so used to people just slamming local models over data issues and I knee jerk replied to you. Sorry.
So no one's using Gemini on HN?
Earlier quoted context omitted.
I think it's decidedly preliminary to compare models using the same .md file, since they respond quite differently to the same input. I try to narrow to the top 2-3 and then refine inputs for each one. For me it's unfortunately not much better than an intuitive process of trial and error. Gemini is not at all unusable. It is quite usable for the tasks it excels at - to the point that it is the top pick for many tasks…
What are your prompting and general tips for using Gemini effectively? And what use cases do you think it’s best suited for?
Somebody else mentioned they had great success at math heavy code. I had to develop a complex piece of software that also integrated into 4 existing systems with a lot of poorly documented constraints. I tried with the major models and Gemini provided the most structured solution that would allow me to work on it, add features etc in the future, and it created an MVP in one shot after working through the planning stage in detail. I have managed to work on that code afterwards quite successfully. It is by far the best model for language tasks like OCR and translation. In my opinion the benchmarks, which put it first for this, are far from showing how far ahead it is because it responds so well to iterating on a prompt. So I think it is good to great for a wide variety of things, but you have to iterate. If what you are doing is simple enough you don't need or want to do that, then use the best model you are already comfortable with. For me that type of work is currently done with GPT 5.5.
Judging by how things are moving ( pricing models, limits, harness patchy updates ), it feels like the real salvation will be a combination of more mature OS models and some open source harness setup like OpenCode or similar. I'm feeling like OS models are nearly there, and with the proper setup and harness might already be there. What are the general thoughts on this ?
This one has "generic healing" for issues present in the current generation of local small LLMs, but if things we see from Frontier LLMs generalize, "optimized healing" for quirks present on your pick of local LLM would be more useful.
Earlier quoted context omitted.
It's well priced but does that have much relevance for "state of the art coding models", specifically? I wouldn't use Gemini 3 Flash or GPT 5.4 mini for anything except the most trivial work, although both are useful for basic exploratory work. So I'm using a heavy model for the bulk of the work and the cost of that so far outweighs the light model that the light model cost is effectively irrelevant.
It's so interesting to see the wild pendulum swings of LLM sentiment here. If one likes a model then it's capable of one-shotting entire apps. Otherwise it's "only suitable for the most trivial tasks". Never in between.
Just last week, I was trying to map the weird and wonderful column names emitted by a NetScaler’s detailed REST log into OpenTelemetry semantics.
The NetScaler is basically abandoned by its dying vendor. Hence, its new features like sending logs directly to Splunk compatible receivers are basically undocumented. I’m sure there’s like three of us masochists out there stumbling our way through the brambles.
The Open Telemetry end is a mess of “deprecated” and “beta”, copying the shifting sands of other cloud native projects like Kubernetes.
Even with carefully curated Markdown documentation references and sample logs, every modern “frontier” AI makes basic mistakes and hallucinates like crazy no matter how I stuff their context.
This isn’t an Erdős problem! This is just getting logs from point A to point B.
It's extra interesting because I think the model people should be talking about is actually not DeepSeek V4 Pro, but the Flash version. When accounting for cache hits, the input price (per OpenRouter) is effectively only 6 cents per million tokens (3 vs 14 cents hit/miss), and 28 cents on output. That's really good efficiency, and it's not a sale price like they are doing with V4 Pro, it's the normal price. It's actu…