dsrtslnd23

8 ms·

91.
by dsrtslnd23 2y ago
same here.
92.
by dsrtslnd23 2y ago
I use o1 with https://aider.chat
93.
by dsrtslnd23 2y ago
are the cover texts LLM created? That seems to be problematic for non-english languages.
94.
Ask HN: Using LLMs for Better Design in Front End Development?
36 points by dsrtslnd23 2y ago|24 comments
95.
by dsrtslnd23 2y ago
is this comparable to asciidoc?
96.
by dsrtslnd23 3y ago
you mean a few hundred Groq accelerators ;-) (they have 230MB SRAM per accelerator)
97.
by dsrtslnd23 3y ago
So unless there are new Croq datacenters coming, this is only interesting for North American users. Otherwise H100 based latency optimized solutions would be faster - in particular for time-to-first-token sensitive applications.
98.
by dsrtslnd23 3y ago
How open is your early access? i.e. likelihood to get API access granted right now
99.
by dsrtslnd23 3y ago
looking at https://github.com/LargeWorldModel/LWM - they seem to indeed suggest to use a TPU vm
100.
by dsrtslnd23 3y ago
Is there any information on a suggested inference setup? I guess they had something different in mind than TPU v4-128 when they put it on HuggingFace?
101.
by dsrtslnd23 3y ago
How can people live without emacs open?
102.
by dsrtslnd23 3y ago
See the conclusion remarks in the paper - they acknowledge that voice cloning is not that good (yet).
103.
Ask HN: Why does Apple want me to expose my iPhone password to repair my screen?
56 points by dsrtslnd23 3y ago|70 comments
104.
by dsrtslnd23 3y ago
It says "Outlines 〰 is compatible with all models.". But does this actually work with gpt3.5-turbo or gpt4? I was using guidance before and you only get value when using davinci due to the constraints of chat api based models.
105.
by dsrtslnd23 3y ago
From the paper it looks like you would get GPT-4 level quality with the 65B model - but if you just do some random tests you will quickly figure out that is not even remotely the case. There must be something seriously wrong with the benchm
106.
by dsrtslnd23 3y ago
no chinese TTS either.
107.
by dsrtslnd23 3y ago
reproducing the english TTS demo resulted in very underwhelming quality.
108.
by dsrtslnd23 3y ago
I tried the english TTS example and the result is quite underwhelming (compared to bark or polly/azure-tts ). It sounds like TTS systems one or two decades ago. Would those language-specific TTS models need to be finetuned?
109.
by dsrtslnd23 3y ago
Anyone with API access tried this for coding already?
110.
by dsrtslnd23 3y ago
Did I get this right that this is just a paid API release - no public code/model available?
111.
by dsrtslnd23 3y ago
Nope. GPT4 is way better from my experience.
112.
by dsrtslnd23 3y ago
Install a local LLM (e.g. Vicuna https://github.com/lm-sys/FastChat ) to have an offline alternative for stackoverflow (and GPT4).
113.
Ask HN: What GPT4 frontend do you use?
5 points by dsrtslnd23 3y ago|5 comments
114.
by dsrtslnd23 3y ago
emacs / org-mode I like the endless features it has, that I can extend its features and it is not a product that will be phased out an some point. On change requests: a more modern lisp dialect would be great
115.
by dsrtslnd23 3y ago
I somehow never get bored. Did not happen to me yet. I can get excited on-demand just thinking by myself about various topics.
116.
by dsrtslnd23 3y ago
I use org-mode to stay organized and actively review it - somewhat in a "Getting Things Done" (GTD) manner. Personally I am very sensitive about external interruptions, in particular sounds / voice and have them all switched
117.
by dsrtslnd23 3y ago
Making the LLM an interchangeable core and build the moat with other technology. If the LLM itself is the core technology - there won't be a moat.
118.
by dsrtslnd23 3y ago
Coding, marketing tweets/posts based on rough ideas, website content rewrites.
119.
by dsrtslnd23 3y ago
It really helps me to overcome inertia and unblocks my mind to get started, even if it not always get things right.