informal007

11 ms·

61.
by informal007 1y ago
Finally finished my ideal bracket and quote code in vimrc In the past, I cannot find a great config for this function, now, I finally finished it by myself with help of Gemini, share to people who need it.
62.
by informal007 2y ago
Some students become better because of LLMs, some become worse. It's like some people learn knowledge by TikTok, some just waste time on it.
63.
by informal007 2y ago
It's like painting made by human, not photo captured by camera.
64.
by informal007 2y ago
> Outputs are evaluated with a scoring rubric by Claude Sonnet 3.7. Will Claude Sonnet 3.7 judger favor himself?
65.
by informal007 2y ago
How much GPU memory are required for inference if it's 10M context?
66.
by informal007 2y ago
It's still BF16 model. Deepseek has proved that fp8 is more cost-effectiveness than fp16, isn't it valid for dozens-B model?
67.
by informal007 2y ago
Some machine learning models are explainable, like tree model, xgboost. Neural network model are hard to explain, especially LLM.
68.
by informal007 2y ago
It's like local Google Drive.
69.
by informal007 2y ago
Sorry for the unrelated comment. It seems that I saw "AI" when I glance "A" in the title, I'm like a LLM trained by news in recently years...
70.
by informal007 2y ago
There isn't a better topic for tech CEOs to focus on, it's pathetic for the world.
71.
by informal007 2y ago
I would like to have Mac as my personal computer and digits as service to run llm.
72.
by informal007 2y ago
On the one hand, Xiaomi is enough popular to do such kind of good thing to attract attention for geeks. On the other hand, this can reduce the risk of fraud and cost of anti-fraud, especially when cost of phone is becoming lower and lower.
73.
by informal007 2y ago
Time is always a better evaluater than anyone in any journal
74.
by informal007 2y ago
In enterprise, there will be long content or document be poured into ChatGPT if there isn't policy limitation from company, which can be a meaning training data. At least, there's possibility these content can be seen by staff in
75.
by informal007 2y ago
This remind me why enterprise don't integrated OpenAI product into existing toolset, trust is root reason. It's hard to provide trust to OpenAI that they won't steal data of enterprise to train next model in a market where co
76.
by informal007 2y ago
The result that MCP server returned will be transfer to MCP host(Claude, IDEs, Tools), there are some privacy issues because the process is automatic after one-time permission provided. For instance, when there is something wrong for MCP ho
77.
by informal007 2y ago
Marginalia Search: a non-profit amateur search engineer
78.
Ask: HN How often do you add or delete a rss resource for you rss reader app?
1 points by informal007 2y ago|2 comments
79.
Ask HN: What are the things you haven't accomplished for 2024?
6 points by informal007 2y ago|20 comments
80.
by informal007 2y ago
For nerd like me, this can reduce the level of awkward.
81.
by informal007 2y ago
Thanks for your detailed reply. [1] does works for me. [2] I’m trying, but the chat room isn’t very active. I think it will take more time to receive an invitation.
82.
by informal007 2y ago
Great gift for YouTubers
83.
by informal007 2y ago
Accurate contents are necessary for professional podcasters, it's hard to ensure the accuracy for AI-Generated contents, it's still a challenging problem.
84.
by informal007 2y ago
The purpose of reading is to obtain values, AI-Generated definitely reduce the density of values.
85.
Ask HN: How could I get an invitation to Lobste.rs?
1 points by informal007 2y ago|2 comments
86.
by informal007 2y ago
flash-attention is a additional and later technique to accelerate computation of attention, That's why it's not the default option for llama.cpp
87.
by informal007 2y ago
So great and funny product!!!
88.
by informal007 2y ago
Totally right, regulatory and liability burden take the huge part of cost. However, Consumers can take some of these responsibilities, That's why the price of service mentioned by OP can be less expensive.
89.
by informal007 2y ago
As IndieHackers continue to grow, an increasing number of skilled marketers will join the platform. Maintaining the original quality will become challenging, and I often feel that it's unavoidable for IndieHackers to evolve into someth
90.
by informal007 2y ago
Yes

More ›