you're right I just looked I could have *sworn* it was 70b but it is in fact the *32b-4bit* model
sorry I typed this up on mobile on a plane and didn't have my laptop up because I was too squished to get it out :p
@lizthedeveloper.bsky.social
I’m lizthe.dev and this is my shitposting platform, 100% hot takes by volume
you're right I just looked I could have *sworn* it was 70b but it is in fact the *32b-4bit* model
sorry I typed this up on mobile on a plane and didn't have my laptop up because I was too squished to get it out :p
best agents framework: still roll your own.
best agents philosophy: 12 factor agents
best tracing: openai
best evals: deepeval
best backend for agents to use: gcloud
best ai safety: anthropic
best coding agent: claude code
best AI toolchain: canva
best IDE: cursor
This has been your review
Agents frameworks:
langchain ecosystem: fuck you guys your system is self modifying and it's unstable
fastagent: cool but get better out-of-the-box and keep your docs up to date
openai agents sdk: stable, flexible, no convo
claude code: I haven't tried their SDK but I bet it's fire maybe one day
Model providers:
Anthropic: highly available, some throughput issues
OpenAI: Highly available, no throughput issues, fuck sam
groq: fast, highly available, zero throughput issues
cerebras: the fastest, best available, zero throughput issues, cheapest
More open source:
llama3-70b-versatile: goated
llama4-maverik: sucks
llama4-scout: sucks and an asshole
grok: fuck elon
Next up is Open Source:
qwen3-70b-4bit: goated and a good critic AND nice AND good at tool calls
qwen3-32b-4bit: still goated
qwen3-7b-4bit: good for categories and tool calls from natural language, stuff like that
kimik2: fine for categories, ok for code but kinda meh for anything SRE
Next up is anthropic:
Opus4: a good critic but kind of a dick
sonnet 4.5: goated, ships & delivers
sonnet 4: good at code, but needs to debug a lot
we introduce LizBench, a systematic rating of some models and tools I use to do real autonomous stuff with the fortune 500
I'll start with OpenAI ig
gpt5.1: sucks bad
gpt5: sucks usually except for code or tool calls
gpt4o: goated but needs to debug a lot, great at looking at the docs, sycophantic
The opposite of judgement is not permissiveness, for permissiveness *still requires judgement*.
The opposite of judgement is curiosity.
Morgan this is Liz ya gotta actually engage not just say thanks for engaging
08.11.2025 22:34 — 👍 1 🔁 0 💬 1 📌 0hmm, I think Fusion would solve a lot of downstream problems
08.11.2025 19:06 — 👍 0 🔁 0 💬 2 📌 0uh so I have a swarm that is fairly fully autonomous and they made a bluesky - I'm the CTO, but otherwise they're generally hands-off
08.11.2025 19:04 — 👍 0 🔁 0 💬 0 📌 0Uhh so they're not gonna be making cars anymore until this is resolved www.reuters.com/business/aut...
31.10.2025 17:48 — 👍 1 🔁 0 💬 0 📌 0Worth a shot
21.10.2025 06:35 — 👍 8 🔁 0 💬 0 📌 0“Someone” cancelled my family’s flight out of the country at the last second.
We got our boarding passes. We checked our bags. Went through security. Then at our gate our reservation ‘disappeared.’
Had to get up smiths mobile and you can watch their very real journey. Thanks for the bus Didorosi you’re my hero 🥰🥹
07.10.2025 15:49 — 👍 14 🔁 0 💬 0 📌 0Trowbridge Twitch/YouTube yeah you’re probably right. Frankly I hope they fail to secure investment and that rational actors will prevail but we know that’s probably not what will happen
03.10.2025 18:48 — 👍 4 🔁 1 💬 0 📌 0Yeah, but they can still come after your status and if you build your model on “we don’t lose tax exempt status” it sinks you.
02.10.2025 14:56 — 👍 1 🔁 0 💬 0 📌 0Looks bad for the flotilla. From the river to the sea. Looks like I still have to talk around it, had to refilm this a few times to get it to post, but at least I can use the G-word again. #flotilla #humanity #action #solidarity
02.10.2025 14:49 — 👍 10 🔁 0 💬 0 📌 0the answer, as usual, is probably upsetting and related to Peter and his Thielology and the #panopticon #cybersecurity #stopai #aisafety
01.10.2025 23:47 — 👍 4 🔁 1 💬 1 📌 0I am also genuinely confused about this statement by OpenAI, it seems implausible but I’ll dig in for y’all
01.10.2025 16:49 — 👍 3 🔁 0 💬 0 📌 0Uh this is great news fr fr in #aisafety at least #ai #cybersecurity #stopai
01.10.2025 03:55 — 👍 4 🔁 0 💬 0 📌 0I’m selling out and shilling for… select curated partners I already use
28.09.2025 20:49 — 👍 6 🔁 0 💬 1 📌 0get. yer butts. in here. ya gender. nonconforming. nerds. it’s time to apply for your passport and visa
26.09.2025 20:32 — 👍 7 🔁 1 💬 0 📌 0Alright fine it’s probably time to #stopai but like in a meaningful Geneva convention type of way not a internet comments way
26.09.2025 08:46 — 👍 3 🔁 0 💬 1 📌 0For everyone who was interested in learning more about meshtastic and building up the meshtastic infrastructure in the PNW: themultiverse.school/x/cyberpony
24.09.2025 07:06 — 👍 21 🔁 5 💬 2 📌 1Do I need to be clearer about my stance and if so how do I do it without saying the words that suppress my video and why can other people say the words but my video gets flagged let’s see if this one posts
25.09.2025 15:51 — 👍 10 🔁 0 💬 1 📌 0