Everlier

8.7K posts

Everlier banner
Everlier

Everlier

@Everlier

Building LLM agents & tools https://t.co/ZAIKtgvw8D - Harbor / Facts / Mi @tryjitera

Katılım Nisan 2010
465 Takip Edilen1.3K Takipçiler
Sabitlenmiş Tweet
Everlier
Everlier@Everlier·
You don't even need Kimi 2.5 for a decent local LLM setup. - llama.cpp - Unsloth's Qwen 3.5 35B A3B with UD Q4 K XL quants - OpenCode - av/harbor It'll take a while to download/install, but otherwise it's something that mid-range hardware (>32GB RAM, ~8GB VRAM) can run today.
Everlier tweet media
Fynn@fynnso

was messing with the OpenAI base URL in Cursor and caught this accounts/anysphere/models/kimi-k2p5-rl-0317-s515-fast so composer 2 is just Kimi K2.5 with RL at least rename the model ID

English
9
6
93
12.5K
Everlier
Everlier@Everlier·
@BlockedPaths @loktar00 I think the trend recently is that if there's too much of it but none at all at the same time - that's fake
English
1
0
2
10
BlockedPath
BlockedPath@BlockedPaths·
@Everlier @loktar00 Half the time you cant tell whats real online anymore, Im constantly thinking something is fake. hahaha
English
1
0
2
8
Everlier
Everlier@Everlier·
Building a tool to help catch stale skills
Everlier tweet media
English
2
1
6
246
Everlier
Everlier@Everlier·
@BlockedPaths @loktar00 The whole ecosystem is too much for years already, I have a feeling some of my childhood memories are being replaced with knowledge of some of the AI projects
English
1
0
2
10
Everlier
Everlier@Everlier·
@Art_If_Ficial I only hear people installing skills, never about people cleaning them up :)
English
1
0
3
30
Everlier
Everlier@Everlier·
@morganlinton It's just going to create a mini universe to harvest some energy from it to proceed with this this problem :)
English
0
0
0
26
Morgan
Morgan@morganlinton·
Uh oh. Grok Build just asked if I'm okay with it building this over the course of multiple years, or if I would like it to finish the build in 6 - 9 months 😳
Morgan tweet media
English
6
1
10
687
am.will
am.will@LLMJunky·
The next BIG Grok model is (partly) done training, and appears to be quite strong. Elon in the past has not sugarcoated Grok models not being the strongest when it comes to coding. This change in tone bodes well, and Cursor's post training will only make it better. I'll be excited to see a new entrant into the competition.
Elon Musk@elonmusk

@beffjezos Our recently completed Grok V9 1.5T run is looking great and that is before Cursor data is added in supplemental training

English
5
0
15
1.5K
Everlier
Everlier@Everlier·
@loktar00 I'll quote you on the launch if you don't mind :)
English
1
0
1
7
Artificially Inclined™
Artificially Inclined™@Art_If_Ficial·
@Everlier @LLMJunky 100% That's why I naturally drift towards the dreamers like @poetengineer__ These types of less-traditional UI/UX explorers, who see a task and don't simply 'build it' they imbue it with a fragment of their soul and make it beautiful as well as highly functional
English
1
0
2
17
Everlier
Everlier@Everlier·
@loktar00 That sounds really cool! I'll appropriate "your agent got a skill issue" from your idea :D
English
1
0
2
17
Everlier
Everlier@Everlier·
@Art_If_Ficial @LLMJunky I think many are clinging to the previous anchors of UI design and information architecture, whereas they are just not true anymore. As well as there's a massive gap in what kind of software is reachable
English
1
0
2
9
Artificially Inclined™
Artificially Inclined™@Art_If_Ficial·
@Everlier @LLMJunky You're right - now it's about finding the perfect combo papers, concepts, abstract flotsam, tiny agents, giant agents.. and see what works best There's never been a better time to run everything through an infinite amount of matrices in short time
English
1
0
0
9
Bullpaid
Bullpaid@bullpaid·
@Everlier You are going too fast , can't catch up everything ...
English
1
0
1
6
Everlier
Everlier@Everlier·
@bullpaid All the markdowning is getting out of hand
English
1
0
1
11
Everlier
Everlier@Everlier·
@PavelSnajdr @ItsAlexhere0 pg is like a Volkswagen Passat B6 there are folks over here that'll try to convince you that sqlite is a great choice :)
English
1
0
1
7
𝘼𝙡𝙚𝙭
𝘼𝙡𝙚𝙭@ItsAlexhere0·
Senior backend interview question: Database CPU hits 95% every single day at 2:43AM. No backups. No scheduled reports. No heavy queries deployed. Where do you start debugging?
English
16
1
19
385
Everlier
Everlier@Everlier·
@Art_If_Ficial @LLMJunky Best of luck! Orchestration is far from solved, lots of people try but noone seem to be able to crack it so far
English
1
0
2
9
Artificially Inclined™
Artificially Inclined™@Art_If_Ficial·
@Everlier @LLMJunky I see.. It's crazy how they can get out of phase I'm dusting off my Garmr project now that all of the SOTA's have CLI/TUI - The orchestration is designed to maintain coherent order. When finished, will drop it as my first official os project.
English
1
0
2
10
Thiago Salvador
Thiago Salvador@bettercallsalva·
@Everlier @subquadratic the 7-point jump from 80.9 to 87.6 on SWE Bench Verified is unusual. either Opus 4.7 closed a real gap or the benchmark is leaking signal at this percentile. independent review of 100 more should settle it. single-benchmark jumps tend to be noisy past a certain percentile.
English
1
0
0
11
Everlier
Everlier@Everlier·
@subquadratic published independent testing results. For context, Opus 4.6 is 80.8% on SWE Bench Verified, Opus 4.5 is 80.9%, Opus 4.7 is 87.6%. They are saying they partnered for independent review of 100 more benchmarks. I'm still on the fence about sparse attention, especially masked one like here, but these numbers and how open they are about independent verification give me some confidence.
Everlier tweet media
English
1
0
2
152

Keşfet

@BlockedPaths @loktar00 @SHELLEYBLEND @Art_If_Ficial @morganlinton @badlogicgames @LLMJunky @poetengineer__