ewoodrich

11 ms·

91.
by ewoodrich 11mo ago
I also wouldn't underestimate Google's ability to nudge regular users towards whichever AI surface they want to promote. My highly non-technical mom recently told me she started using Google's "AI Mode" or whatever
92.
by ewoodrich 11mo ago
Wait, are you really suggesting it's somehow an emergent property of any LLM that it will spontaneously begin to praise its largest shareholders to the point of absurdity? Does LLaMA with the slightest nudging announce that Zuckerberg
93.
by ewoodrich 11mo ago
I really need something like this up for tasks I want Claude to run before handing off a task to me as "complete". It routinely ignores my instructions of checklist items that need to be satisfied to be considered successful. I ha
94.
by ewoodrich 11mo ago
But it doesn't evaluate the area that I am most eager to see improvements in LLM agent performance: unattended complex tasks that require adapting to unexpected challenges, problem solving and ambiguity for a long duration without a hu
95.
by ewoodrich 11mo ago
I was pretty disappointed to learn that the METR metric isn't actually evaluating a model's ability to complete long duration tasks. They're using the estimated time a human would take on a given task. But it did explain my i
96.
by ewoodrich 11mo ago
Android permissions began to ask for individual confirmation on first use in Android 6.0 (released in 2015) so the grant-all-on-install model hasn't been how it works in a very long time. Also your narrative about iOS moving from locke
97.
by ewoodrich 11mo ago
> It's wild that most developers in a corporate environment pay Microsoft to use Chat GPT, Gemini, and Claude It definitely didn't help back when my manager asked me to recommend a subscription to buy everyone on our team that
98.
by ewoodrich 11mo ago
Agreed, some days I code for 4-6 hours with agentic tools but 2025 or not I still can't stomach using any of the big three LLMs for all but the most superficial research questions (and I currently pay/get access to all three pai
99.
by ewoodrich 11mo ago
Reminds me of the classic in the genre: how about a couple of weeks of gratitude for magic intelligence in the sky, and then you can have more toys soon? Sam Altman, Sep 12, 2024 https://xcancel.com/sama/statu
100.
by ewoodrich 11mo ago
I can sympathize with being mistakingly accused of using LLM output, but as a reader the above format of "Its not x - it's y" repeated multiple times for artificial dramatic emphasis to make a pretty mundane point that could
101.
by ewoodrich 11mo ago
The very light moderation (that even shows dead comments from banned accounts) and clean, minimal frontend with essentially no restrictions on creating throwaways also makes it pretty attractive for "my first AI app" experiments.
102.
by ewoodrich 11mo ago
> The last point is how it usually fails in my testing, fwiw. It usually ends up borking something up, and rather than back out and fix it, it does a 'git restore' on the file - wiping out thousands of lines of unrelated, unsta
103.
by ewoodrich 11mo ago
Could you give an example of what they're gatekeeping for Copilot exclusively? I'm kinda confused because Copilot in VS Code isn't exactly a powerhouse of unique features in my experience, it still feels well behind Roo/
104.
by ewoodrich 11mo ago
What exactly are the "extras" that Copilot gets? I've had a Github Copilot subscription from work for 1yr+ and switch between the official Copilot and Roo/Kilo Code from time to time. The official Copilot extension has i
105.
by ewoodrich 11mo ago
Thanks! I'll give that a try.
106.
by ewoodrich 11mo ago
I didn’t say it was invalid, just that it was needless. When I bought the laptop Gatekeeper was a tolerable nuisance and I was fine with the tradeoff given the security benefits. The removal of the hotkey (which also required changing a set
107.
by ewoodrich 11mo ago
I ran into this exact same thing recently with CSVs downloaded from my own app. I tried a few different filetypes and was baffled how seemingly any filetype I downloaded triggered Gatekeeper regardless of the app I set to open it (including
108.
by ewoodrich 11mo ago
It really bothers me that Apple removed any convenient shortcut to bypass Gatekeeper like the old Control-click [1] hotkey. Apple's relentless ratcheting of the difficulty/annoyance of Gatekeeper has just about pushed me over th
109.
by ewoodrich 11mo ago
I've only had that happen when I use /compact, so I just avoid compacting altogether on Codex/Claude. No great loss and I'm extremely skeptical anyway that the compacted summary will actually distill the specific actiona
110.
by ewoodrich 11mo ago
Dispensaries in OR/WA love $2 bills, for some chains they're as unremarkable as a $1 and must special request them in bulk to keep on hand for making change.
111.
by ewoodrich 11mo ago
> The text sections take implementation details that don't matter and present them to the user like they need to know them. Yeah this seems to be a recurring issue on each of the repos I've tried. Some occasionally useful table
112.
by ewoodrich 11mo ago
Hah I misinterpreted it a different way as a kid, for a long time I thought it was like a collective delusion where the shared experience of contributing insubstantial garnishes to a pot of water tricked everyone into finding it filling and
113.
by ewoodrich 1y ago
I set up a URL redirect rule in Edge/Brave/Chrome with the extension URL Auto Redirector (previously used Redirector but it was removed, there are other alternatives available for Firefox I'm sure). I also found a similar fro
114.
by ewoodrich 1y ago
Also there is a fixed price Claude Code plan for GLM 4.6 from z.ai, I pay for the cheapest ($6/mo) as an alternate/fallback for Claude Code and Codex. I've been surprised by how similar in capabilities all three of them are,
115.
by ewoodrich 1y ago
There is now also copilot-cli that's a clone of Claude Code and by default runs with Sonnet 4.5. I haven't spent much time with it on really complex stuff yet but it's a nice option to have available if you have a Copilot Bus
116.
by ewoodrich 1y ago
The weak points raised by the parent comment are specifically examples where the problem exists outside the model's "peripheral vision" from its context window and speaking from personal experience, aren't as simple as a
117.
by ewoodrich 1y ago
It was certainly jarring reading that immediately after accusing someone else of having never played the games. (Who is clearly playfully poking fun at something he enjoyed playing but can recognize as being constrained by computers of the
118.
by ewoodrich 1y ago
It’s less rigid than a command line but much less predictable than either a CLI or a GUI, with the slightest variation in phrasing sometimes producing very different results even on the same model. Particularly when you throw in agentic c
119.
by ewoodrich 1y ago
I'm struggling to think of a reason why being "treated as an always-idiot" is an actual negative in this specific example. I use Bitwarden and when the password autofill doesn't work as expected my first assumption from
120.
by ewoodrich 1y ago
> aren’t yet seeming to train on how to build maintainable software. A while ago I discovered that Claude, left to its own devices, has been doing the LLM equivalent of Ctrl-C/Ctrl-V for almost every component it's created in a

More ›