blixt

10 ms·

91.
by blixt 2y ago
LM Studio has been around for a long time and does a lot of similar things but with a more UI-based approach. I used to use it before Ollama, and seems it's still going strong. https://lmstudio.ai/
92.
by blixt 2y ago
I agree. In theory you need to just distribute serialized patches but in a real world backend scenario you may need to integrate with knowledge of current document state, user identity, and possibly even some level of access control. I’ve w
93.
by blixt 2y ago
Pretty interesting watching their tech explainers on YouTube about the changes in their AI solutions. Apparently they switched from CNNs to transformers for upscaling (with ray tracing support) if I understood correctly though for frame gen
94.
by blixt 2y ago
Amanda Askell at Anthropic works on Claude's personality and shows off how it's been finetuned to have certain qualities that makes it seem more relatable with a personality. It's important to remember that a lot of this beha
95.
by blixt 2y ago
Lovely ending, and I appreciate how short this one is. For me it really does induce some mixed feelings for what we did to the web while at the same time I really enjoyed the nostalgia. Another game I sunk way too much time into to get to t
96.
by blixt 2y ago
These results are fantastic. Claude 3.5 and o1 are already good enough to provide value, so I can't wait to see how o3 performs comparatively in real-world scenarios. But I gotta say, we must be saturating just about any zero-shot reas
97.
by blixt 2y ago
I was hoping to find a part 2, but since part 1 was written over 2 years ago I guess there's not much chance of that...
98.
by blixt 2y ago
Yeah that's fair, I think everyone has their own learning style. I mostly felt a need to respond with a slightly more optimistic view on what this new technology means for juniors, in particular responding to the part that "[AI co
99.
by blixt 2y ago
I see the same things as Addy, though I'm not 100% sure it's something new happening because of AI assistants. I started learning programming in the late nineties as a 9-year-old sitting at a library paying 10 NOK for an hour of i
100.
by blixt 2y ago
Seems somewhat likely to me. They probably even trained a model to do both frame generation and upscaling to allow the hardware to work more efficiently while being able to predict the future based on user input (to reduce perceived latency
101.
by blixt 2y ago
This is quite beautiful. It hides several ugly warts of Go without necessarily adding more complexity. Result<T> instead of (T, error) and ? for propagation is just so nice. Structured enums is something I find myself needing all the
102.
by blixt 2y ago
I can’t say I share your optimism. I’ve seen plenty of developers point to graphs like these as a reason for why they picked a language or framework for a problem. And it comes down to the benchmark how good of a proxy it actually is for su
103.
by blixt 2y ago
While it’s nice to compare languages with simple idiomatic code I think it’s unfair to developers to show them the performance of an entirely empty function body and graphs with bars that focus on only one variable. It paints a picture that
104.
by blixt 2y ago
Took some time to read the linked Tailscale blog posts as examples of horrible hacking around Go. And I would agree it's crazy for a product oriented startup to do this, but this team is also made up of people that have contributed a l
105.
by blixt 2y ago
I feel like the author is fed up with being disagreed with, yet makes very opinionated statements about how others should (or will, at some point in the future) feel. They make good points about problems in Go, but I gotta say they also mak
106.
by blixt 2y ago
Really interesting findings around fine-tuning. Goes to show it doesn't really affect the deeper "functionality" of the LLM (if you think of the LLM running a set of small functions on very high-dimensional numbers to produce
107.
by blixt 2y ago
I don't completely disagree but I believe it's a bit more fuzzy than that. From what I understand, the models learn a very compressed version of what they receive as input and produce as output. While not sufficient to generalize,
108.
by blixt 2y ago
I think it's infeasible to train on bytes unfortunately, but yeah it also seems very wrong to use a handwritten and ultimately human version of tokens (if you take a look at the tokenizers out there you'll find fun things like reg
109.
by blixt 2y ago
Will you come back to Europe?
110.
by blixt 2y ago
My partner has ~300 tabs open at any point in time in Safari for iOS and uses the tab grid as a kind of visual registry of things to read, interesting events/restaurants, in-progress work, etc. It makes me panic looking at it but I thi
111.
by blixt 2y ago
They say Hertz is first of its kind but Moshi is another duplex audio model from earlier this year that seems to perform similarly (and it runs on a MacBook): https://github.com/kyutai-labs/moshi
112.
by blixt 2y ago
Gotta say I was confused for a second but yeah apparently si.inc and ssi.inc are the domains for two different AGI companies and I can only assume it’s intentional?
113.
by blixt 2y ago
Good points, and I agree. I think what bothers me is that a lot of the advice you might read out there is biased towards what you should build if you already had hundreds of employees and millions of users. Sprinkle in a bit of incentives f
114.
by blixt 2y ago
I agree with the general message of this article but I want to rant a bit on what exactly is "boring architecture". I haven't quite managed to put this into succinct words, but in my experience building the innards of both ba
115.
by blixt 2y ago
Yeah probably, it remains to be seen if these models can actually help guide a live session towards the goal. At least it's been shown that these types of world models can help a model become better at achieving a goal, in lieu of a ha
116.
by blixt 2y ago
I guess it comes down to your definition of state. I'm not saying there's enough state for this to be playable, but there is clearly state and I think it's important to point out how impressive the amount of temporal consiste
117.
by blixt 2y ago
There's plenty of evidence of state, just a very short-term memory. Examples: - The inventory bar is mostly consistent throughout the play - State transitions in response to key presses - Block breakage over time is mostly consistent -
118.
by blixt 2y ago
Super cool, and really nice to see the continuous rapid progress of these models! I have to wonder how long-term state (building a base and coming back later) as well as potentially guided state (e.g. game rules that are enforced in traditi
119.
by blixt 2y ago
It is a proxy to questions that pop up as part of other sessions in my daily use. The LLM chooses its query itself and as long as it is not fine tuned to avoid the “listicles”, or its underlying search engine are not putting importance on m
120.
by blixt 2y ago
It would be very odd if I have to speak to an LLM the same way I “speak” to Google. The point of this whole feature seems to be that the LLM chooses when and how to search to best respond to the user.

More ›