kmacdough

8 ms·

91.
by kmacdough 1y ago
This kind of underscores the author's point, though. These books in question give the impression that they supply concrete business advice, but much of it is highly non-transferrable. Or at least requires significant discipline to sepa
92.
by kmacdough 1y ago
> I was hoping for something more like "the problem is specified such that invalid solutions aren't even representable, so only acceptable solutions are considered." How on earth would one come up with a model where "
93.
by kmacdough 1y ago
Yes, because this is precisely what rust has proven it is exceptional for: tooling that needs to be very correct and strives to be fast.
94.
by kmacdough 1y ago
I certainly agree in many cases like those you mention, particularly for unprompted popovers like notifications. For desktop professional software, on the other hand, I believe the value of swift muscle memory supercedes this. In this case,
95.
by kmacdough 1y ago
And to those who "say" at least now they're honest, I say "WHY?!" Unconditionally being "good" would be better than disguising selfishness as good. But that's not really a thing. Having to maintain th
96.
by kmacdough 1y ago
>In the big picture, I think there aren't many hockey sticks/exponentials left in LLMs to discover. That is not true about other AI/ML. We do appear to be hitting a cap on the current generation of auto-regressive LLMs, bu
97.
by kmacdough 1y ago
I would be extremely hesitant to assume a direct relationship between pricing and cost. A behemoth like Google is very willing to take significant losses for years to grow market share. Back in 2014-2015 Uber often charged less than the Bos
98.
by kmacdough 1y ago
Keep in mind this release was never intended to prove superiority. Rather, it shows an alternative structure with some promising performance characteristics. More work needs to be done to show real application, but this very valuable learni
99.
by kmacdough 1y ago
What are the barriers to mixed architecture models? Models which could seamlessly pass from autoregressive to diffusion, etc. Humans can integrate multiple sensory processing centers and multiple modes of thought all at once. It's bak
100.
by kmacdough 1y ago
It appears to be purpose-trained for coding. They also have a generalist model, but that's not the one being compared. I agree, the comparison is dated, cherry-picked and doesn't reference the thinking models people do use for cod
101.
by kmacdough 1y ago
This isn't a fix, but a small patch over a much bigger issue: what increases temporary engagement and momentary satisfaction ("thumbs up") probably isn't that coupled to value. Much like Google learned that NOT returning
102.
by kmacdough 1y ago
I'd like to see OpenAI and others get at the core of the issue: Goodhart's law. "When a measure becomes a target, it ceases to be a good measure." It's an incredible challenge in a normal company, but AI learns and
103.
by kmacdough 1y ago
Here's a paper on agent architectures including multi agent. A bit old at this point, but a good overview. https://arxiv.org/abs/2404.11584
104.
by kmacdough 1y ago
These ensembles have been tested throughout AI progress. Well scaffolded larger models have historically come out ahead in both quality and speed/cost. Perhaps this is a parricularly effective ensemble, but I would need to see real da
105.
by kmacdough 1y ago
Bookmarks are more permanent. Open tabs for me are a physical representation of a task in process. Tabs for a purchase show what needs consideration. Could be options, reviews, learning materials. For a project, tabs might include specific
106.
by kmacdough 1y ago
> And then if this becomes more widespread an the existing software companies that do business with government will start sucking money out of such grants and the government quality code. Agreed, but keep in mind this must be compared to
107.
by kmacdough 1y ago
This is an argument in favor of all anti-competitive walled garden moves. But we've seen time and again the degrading service and price gouging that ultimately comes. People have a right to use the things they own as they please. Compa
108.
by kmacdough 2y ago
Look up Claude Code, Cursor, Aider and VSCode's agent integration. Generally, tools to use AI more actively for development. There are others as well. Plenty of info around. Here's not the place for a tutorial.
109.
by kmacdough 2y ago
It's such a rapidly developing field with much of the progress happening in small labs on the open source models. Eventually, the field will coverage and stabilize. For now, the bet is too be open and supportive, to be close to the pro
110.
by kmacdough 2y ago
Sure, spotify has maintained a near-complete catalogue where Netflix hasnt. But I no longer find Spotify any good at finding new music, beyond manually looking through artist catalogues. For context, try out Pandora's recommendations.
111.
by kmacdough 2y ago
It's not blame, but it's a striking reality that needs to be kept at the forefront. It introduces a substantial set of novel failure modes, like cross-tool shadowing, which aren't obvious to most folks. Making use of any exte
112.
by kmacdough 2y ago
> I'm wondering if it is something like they can track mouse movement Yes, this is a big signal they use. > adding some more human like noise to the mouse Yes, this is a standard avoidance strategy. Easier said than done. For eve
113.
by kmacdough 2y ago
> It's impossible. How do you even define advertising? Seems you're insistent on letting the perfect be the enemy of the good. This is precisely why judges exist. It would not be difficult to define Ads well enough to cover 99%
114.
by kmacdough 2y ago
Computers can and should help along the way, but Dijkstra's argument is that a) much of the challenge of human ideas is discovered in the act of converting from natural to formal language and b) that this act, in and of itself, is what
115.
by kmacdough 2y ago
I am having a hard time finding controlled testing, but the premise is straightforward: different modalities encourage different skills and understandings. Text builds up more formal idea tokenization and strengthens logic/reasoning wh
116.
by kmacdough 2y ago
Vague question, as privacy means different things to different people, but its very hackable, so as long as you're running Android you should be able to make it work how you want. By default it uses the Pebble App for sync. You can dec
117.
by kmacdough 2y ago
It's much more generic than that. It's highly flexible and intuitive hierarchical information organization software. At least that's how I'd describe it. The innovation isn't the organized structures themselves eith
118.
by kmacdough 2y ago
That or it's the luckiest coincidence! In all seriousness, Apple is fairly consistent about not pushing specs that don't matter and >256GB is just unnecessary for most other common workloads. Factors like memory bandwidth, core
119.
by kmacdough 2y ago
I understand the principles of MOE, but clearly not enough to make full sense of this. Does each expert within R1 have 37B parameters? If so, is QwQ only truly competing against one expert in this particular benchmark? Generally I don'
120.
by kmacdough 2y ago
I think they're trying to make the distinction between "we sell your searches and clicks attached to your personal id" and "we sell derivative aggregated information like we have a lot of users of X style to advertise to

More ›