@snellingio
 Oklahoma City
Joined July 2009
  • Tweets3.3K
  • Following772
  • Followers1.3K
321Media
Far too many people: "Benchmarks are bad. I prefer [thing correlated at 0.9 with Benchmarks]."
1
307
This seems bad.
104
1,853
131
38,500
739,777
Sam Snelling retweeted
Florian Brand
@xeophonOct 21
Oh, you aren’t updating your dependencies every hour? Well, too bad, you are vulnerable Oh, you are updating your dependencies every hour? Well, too bad, you are somehow also vulnerable
JFrog Security
@JFrogSecurityOct 21
🚨 JFrog Security Research has identified a supply-chain attack involving malicious code in the LiteLLM 1.107.0 PyPI release. Avoid this version and investigate environments where it was installed or executed. More details will be shared soon in our blog post.
35
231
14
4,048
78,205
Sam Snelling retweeted
lauren
@potetoOct 21
every security engineer this week
60
98
15
1,813
53,018
Well hot damn maybe there is hope for local inference.
Richard Socher
@RichardSocherOct 21
Replying to @RichardSocher
Flint is the best-performing model in its class (MoE models under 50B) on agentic coding benchmarks like SWE-bench Pro v2, Terminal-Bench 2.1, and SWE-bench Verified. It also rivals similarly sized dense models that are much slower to run.
2
10
464
Sam Snelling retweeted
Amp
@AmpCodeOct 20
21 people, a single meeting a week, no quarterly goals, no pull requests, no mandatory code review, no titles (except one dictator), a hundred releases a day, bugs fixed in minutes, famous for killing features. @sqs and @thorstenball on how the Amp team works:
17
42
15
523
362,553
yo this is brutal lol
Marques Brownlee
@MKBHDOct 21
The current state of apps on iPhone Duo: youtu.be/ESABTN4VtrA
56
yeah I actually think this is awesome? we're in the era of disposable software even more than personal software. time to decide if something is worth your attention has never been lower.
Jack Ellis
@JackEllisOct 20
Last night I asked ChatGPT to build a replacement for an app I pay $10/month for. I wanted to build out some features specifically for my use case. Today I woke up and checked it, it was the perfect clone. Then I decided that this project was a waste of my time and deleted it.
1
91
Sam Snelling retweeted
AI at Meta
@AIatMetaOct 19
Today we published a technical report on Muse Glimmer, our 30B open-weight model built for agentic tasks on consumer hardware. It runs at practical speeds on a Mac or PC with a single consumer GPU at 4-bit precision, and leads comparably-sized open models on tool use and multi-step planning. Read about the model architecture, distillation recipe, quantization and more: go.meta.me/4Bcvxol
160
187
143
1,381
217,473
Sam Snelling retweeted
Thariq
@trq212Oct 19
I think most people multi-tasking with agents could benefit from slowing down most of the stuff you're doing wont really matter- what's your P0? what would it look like if you gave it your full attention?
563
379
179
6,895
451,062
okay what you need to build gameboy games with a real cart: krikzz everdrive x5 microsd card + sd adapter gameboy color or equiv (chromatic, etc) workflow is -> work on the .gb game locally, then move to microsd card, put microsd card in x5, and play! I thought it was a lot more complicated, but it is not at all.
1
1
162
Sam Snelling retweeted
will brown
@willcbOct 18
Replying to @scaling01
i am explicitly factoring that in btw that’s just not how GDP works
3
3
113
5,914
My guess is Luna is super small. It could be wishful thinking though. I’d love to know the size of Luna and Haiku.
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)
@teortaxesTexOct 18
If you think that DSV4.1 to V4.1-1017 improvement is solely due to OpenAI distillation, then I wonder why OpenAI couldn't distill a Luna that beats even V4.1. Just how tiny must Luna be then? and how high their margins, given that it's slower too, despite superior hardware?
78
Sam Snelling retweeted
DeepSeek
@deepseek_aiOct 17
🚀 DeepSeek-V4.1-Flash-1017 is here! Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API. 🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, WorkBuddy, and more. Upgrade to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model. 1/4
1,617
4,129
2,835
37,091
7,946,754
missed opportunity
49
homie gonna carry them through ipo at this rate
kitze 🛠️ tinkerer.club
@thekitzeOct 15
went from 2 claude accs this morning to 4 now 💀 5th soon also canceled the $500/mo copenai sub cuz i think to charge that amount / mo for something it has to be EXTRAORDINARY and right now their entire offering is the opposite of that 💀
74
Sam Snelling retweeted
Nathan Lambert
@natolambertOct 15
Guys this is like my superbowl I waited so long for US open weight models to be relevant again. Before today the US models were still behind Chinese models from as early as Feb. 2026. Still more work to do! This cut the gap by at least half, in time (say 3 months)
57
49
10
642
58,241
the claude desktop app went from the absolute shittiest way to use claude, to one of the best ai apps I've used. I think it's mostly a speed / responsiveness thing they've really improved. Not really a magic sauce or anything
59
This was undersold imo. Tab completions are back and they are usually pretty good. I have a > 50% hit rate
Tibo
@thsottiauxOct 9
Day 5/ Composer predictions in the desktop app. Often leads to a double take with how on point they are. Included in the Pro plans without consuming usage.
1
57
Sam Snelling retweeted
Daanish Khazi
@bertgodelOct 14
We are releasing Ultramarine 1 Flash, a frontier-class model built for knowledge work. It is open-source and achieves similar performance to Fable 5, while being 60x cheaper. It was post-trained by a team of 5 for ~$300k all-in. Available on API now, with weights releasing in 2 weeks.
606
694
376
8,102
1,727,563
Hmm, I guess we’ll see bc I just noticed I had 70gb of worktrees in the .codex folder for a single project 😵‍💫
Tibo
@thsottiauxOct 13
Day 9/ We've rolled out optimizations to how we store state on disk for the chatgpt & codex desktop app that should reduce the amount stored on disk by 60%. Small ship but we've heard enough feedback about power users running out of disk space.
2
1
134
turns out that just because it shows all the work trees, it can't actually delete them all. Had to manually go in and remove them
15
Load more