topmass retweeted
We're giving away a Hogwarts Legacy custom wrapped NVIDIA GeForce RTX 5090 to celebrate its official release with DLSS 5 🪄
To enter:
🟢 Share this post
🟢 Comment #RTXON
Terms and Conditions: nvidia.com/en-us/geforce/con…
topmass retweeted
Shout to @topmass from my AI Discord. He’s working on creating the battle between Sam Altman and Dario we really want to see 😂
This is locally generated with open source local AI gen models too 🤯
topmass retweeted
Replying to @maximelabonne
Was a good idea
Look at it like this
Bad guys have it (but they would have had it anyway)
Good guys with lots of money already have Mythos, Daybreak
Now, good guys without lots of money have an option to defend themselves without hiring a security firm.
topmass retweeted
Introducing Bot Mode for Hermes Desktop.
Your agent profiles become a series of named Bots. Each Bot has its own role, model, memory, skills and profile picture; Bots can use any model and even communicate with each other.
Build a specialist Bot once to use it forever.
Linux adoption among programmers is about to go parabolic. Any missing app can be recreated easily. The open-source advantage with agents is unstoppable. Nobody is going to be waiting for Apple to sign their shit. Just keep a Mac Mini as a remote builder for iOS work. Done.
topmass retweeted
this take is stuck in 2025.
codemode solves most of the issues outlined and also discounts nice properties of mcp like elicitation.
how does your agent learn about all the clis it has available? skills? their descriptions will do the same to context as poorly designed mcps
clis also require a secure execution environment where the agent is running as opposed to remotely in secure isolates
please stop giving your agents MCPs and give them CLIs instead 😭🙏
you can load up an agent with 100 cli tools but you can’t load 5 mcps before it shits the bed with context rot
all models at this stage are already trained to master terminal commands
literally give them cli tools with a help wizard and your agents will learn how to use it in the moment
all you have to do is tell them what tools they have access to and when to apply them in their workflow and they will be proficient by nature
i am incredibly bearish on MCPs and never use them because they’re extremely inefficient
you can’t even fit everything needed for an api in most integrations without a minimum of 5 tools lmfao
the tech needed is the same as well, both MCPs and CLIs need a host server invoked when called
security wise they’re also fundamentally the same
for use in applications, eg. apps that aren’t computer harnesses like claude code or codex (which both actually use the terminal)
i’ll give the agent one native function call which represents the terminal
and it calls the tool with the arguments it needs every time without fail
you don’t even need a separate host when it’s in your app, you can just parse the function call directly in backend logic
just make sure to replicate the terminal experience 1:1
also note that the cli output also act as a prompt itself
outputs should contain the result it needs
now it’s possible for the agent to use the tool wrong or use invalid arguments
when this happens the ERROR OUTPUT is also a prompt
if the error output explains the problem and how to correct it, the agent will self correct on the spot
try this and save your tokens !!!
topmass retweeted
This is going to have an opposite effect that the decels want. It's a huge open source AI accelerant, an accelerant for corporations, and enterprises
Now it's a real race. You either make your own AI infrastructure or you don't have a seat at the table
Introducing OpenQuick - Fully open source variant of shopifys Quick super configurable - easy to deploy and runs each quick page on a single durable object - thanks @CloudflareDev
Check it out ⬇️
topmass retweeted
4-bit Qwen3.6 MTP GGUF managed to search 70+ sites from a single prompt.
Try this locally on 20GB RAM via Unsloth Studio.
Unsloth now supports auto MTP + speculative decoding & auto-selects the best MTP settings for your device (Mac, CPU, GPU).
GitHub: github.com/unslothai/unsloth
topmass retweeted
Crazy how backwards this is. The "yes guy" is now in infinite supply, just type `claude` and you've got a new one. The "no guy" is the one in demand: the one who points out all the problems with all the slop your slopcannons are trying to merge to prod.
This quoted post is unavailable.
Replying to @sairahul1
nvidia did the same with one of the top leaderboard models right now that's literally 0.6b params. if you're using a STT model thats larger than 0.2-3b params it's a waste of your computer. huggingface ASR leaderboard has the best small options but for now just use parakeet.
topmass retweeted
Replying to @thegeeknarrator
I disagree. Code is slow for humans. The more we read or write it the slower we go. To gain productivity from AI we need to disengage from code and put our energies into managing the structure, not the syntax, of the code.
topmass retweeted
Replying to @wookash_podcast
I don’t review code written by agents. I measure things like test coverage, dependency structure, cyclomatic complexity, module sizes, mutation testing, etc.
Much can be inferred about the quality of the code from those metrics. The code itself I leave to the AI.
Humans are slow at code. To get productivity we humans need to disengage from code and manage from a higher level.
switch to linux. wsl works but you'll spend more time debugging windows quirks than actually running models. at 2am when something breaks you want it to just work. ubuntu on a 3090 with llama.cpp and hermes agent just works. make the switch and never look back.
This Post is from an account that no longer exists. Learn more
topmass retweeted
I'm not joking and this isn't funny. We have been trying to build distributed agent orchestrators at Google since last year. There are various options, not everyone is aligned... I gave Claude Code a description of the problem, it generated what we built last year in an hour.
topmass retweeted
I want to take Anthropic's word here, but it's genuinely hard to do so. They have offered no transparency into what caused these outages, and they didn't even acknowledge them until days after they were resolved.
If you want us to take you at your word, you need to tell us what happened. We cannot trust a company that doesn't disclose why their product didn't work properly for months at a time.
Replying to @claudeai
We’d also like to address a concern we've heard in the community: We never intentionally degrade model quality as a result of demand or other factors.
























