@josevalim Better than 4.6, substantially better than 4.1 judging by the amount of times i have to stop and yell at it. Note I have not been using it with elixir
English
Łukasz Samson
231 posts
I have to say I barely noticed a difference between Anthropic models in my workflows since 4.1 days but Opus 4.7 is clearly worse to me.
It just feels… lazy?
I have to frequently remind it to consider additional scenarios and conditions, which I never had to in the past?
English
@ibuildthecloud No. LSP APIs are all cursor position oriented, not natural for LLMs. Spoken as a long time LSP maintainer
English
Has anyone gotten an agent to do anything interesting by calling an LSP?
English
@local0ptimist @gfodor In my experience codex looses track more often than opus
English
A little early to tell but compaction seems close to solved now. Wonder if anyone who's Wiggum'ed much can comment on it.
gfodor.id@gfodor
this is probably part of the reason the tone changed in the last few weeks by the heads of the labs wrt acceleration. the goal of an AI researcher by 2027-2028 seems like it's gonna end up too conservative. flywheel is spinning. whatever they're doing for RL with 5.4 is nutty.
English
@bcardarella Reminds me of Vala, C# clone that translates to C with gobject. My team once started a project using it but after some time we rewrote everything in raw C
English
I have what may turn out to be a bad idea, but I might spike it out. An Elixir-like strongly type syntax that lowers to Zig. Just FP, no processes, actors, VM, or other BEAM stuff.
English
@akoutmos @josevalim @dashbit You may find my newly released library interesting x.com/lukaszsamson/s…
Łukasz Samson@lukaszsamson
I've just released spark_ex - a native elixir Apache Spark Connect client with Livebook integration
English
@josevalim and the team over @dashbit have done an amazing job with the Numerical Elixir libraries. If you are not using libraries like Scholar/Explorer for your analytics needs you're missing out!
That being said...making good progress on the Elixir+Finance book this morning 😉
#ElixirLang
English
Check out a demo notebook github.com/lukaszsamson/s…
spark_ex docs
hexdocs.pm/spark_ex/readm…
English
I've just released spark_ex - a native elixir Apache Spark Connect client with Livebook integration
English
@whatyouhide Let it crash was slop engineering even before vibe coding took off
English
I never shared this view of "let it crash". Things will NOT resume instantly with a clean state. Most of the time, at least in my experience, things will resume instantly but in the same broken state + errors will be a nightmare to debug.
Shannon Sands@max_paperclips
The only problem with LLMs writing Elixir, is being Python-maxxed, they are SUPER defensive. "No, just let the process crash, it'll resume instantly with a clean state & recover, it's fine" is just incredibly confusing to them
English
@fbettag Been there. IO and concurrency are great but decimal arithmetic is painfully slow on BEAM and you pay the price of GC due to immutable data structures and limited true O(n)
English
With a little claude help I ported karpathy's microchat to elixir. Surprisingly it's faster than python
English
@thorstenball LSP is a yesterday tech. The protocol is cursor position oriented and not suited for agents. I’m speaking that as maintainer of an LS
English
Exercise:
Write a small agent, give it that tool, see how often the model uses it. Then write a furious system prompt. Then see how often the model uses it. Then you're enlightened.
English
@trq212 I noticed significant quality drop after CC upgrade to v2.72
English
We've received some feedback about a potential degradation of Opus 4.5 specifically in Claude Code.
We're taking this seriously: we're going through every line of code changed and monitoring closely.
In the meantime please submit any transcripts with issues through /feedback
English
@alkadaemon @ZachSDaniel1 The plan is to support ElixirLS until Expert reaches feature parity
English
@ZachSDaniel1 So we should use this instead of ElixirLs on vs code? Why?
English
Expert LS is out yall. #ElixirLang github.com/elixir-lang/ex…
English
Github Actions: proudly wrapping AWS CLI one liners into fullblown nodejs apps
English
@josevalim There is ast-grep tool and an MCP wrapper github.com/ast-grep/ast-g…. I have it and claude uses it from time to time. Much less than cat & grep
English
Anyone exploring query expansion for RAG in codebases?
For example, instead of doing a vector search, you could instruct the LLM to expand the query whenever searching. E.g. to find the authentication code, it could query for "authentication|auth|login|signin|credentials".
English
@ryanwinchester Still upset at phoenix for changing the frontend every minor release making the generators useless in any longer lived project
English
ElixirLS v0.28.0 released
Highlights:
Added support for OTP 28 on elixir 1.18.4
Support for go to definition and find references in multialias directives
ElixirLS will now try to autodetect path to elixir standard library
English
@dsincl12 @codestirring There’s a Shadcn clone SaladUI salad-storybook.fly.dev/welcome though it’s not feature complete yet
English
@codestirring +1 on DaisyUI too much and too unprofessional IMO. I understand why but I would prefer something minimal instead and I think Tailwind is enough as well. We need Shadcn but for Phoenix.
English
Started building a new app tonight using the latest Phoenix RC.
I’m liking things like Scope and I thought I’d be really up for DaisyUI but something about it isn’t clicking - not in love with a lot of their components.
Anyone else?
#myelixirstatus
English