Nif A

All opinions are mine and not generated by AI 🤖 |💻| Agilent | Ex-Cerebras | CMU

[Nif A @niferential]

Hey I’m Nifasath!

> Prev @cmu @cerebras @agilent @ibm

> Backend & ML Infra Engineer with experience building AI systems around messy workflows, sprawling data, and mission critical infrastructure

> Looking for my next role - DMs open, intros appreciated

[Nif A @niferential]

apparently a poem I wrote in 5th grade about my grandmother is 91 % ai generated text - id rhymed funny with runny smh

[Nif A @niferential]

@mewwts As a former extremely curious 3.5 year old what helped keep that curiosity alive was being told why and making sure I actually understood

[Nif A @niferential]

@andruyeung they definitely downgraded the models the agents are running on and it shows

[Nif A @niferential]

@strimblez between a star vs trying to cover up an angry pimple its the easy pick

[Nif A @niferential]

is my muse just 3 call center employees in a trench coat?

[Nif A @niferential]

@krispuckett im fatigued by older models disappearing

[Nif A @niferential]

@jjacky is that similar to artificial sugar - it wasnt too bad

[Nif A @niferential]

til i learned there are different types of wd-40

[Nif A @niferential]

@thekitze shut it down. i spin up a lot of temporary bots for timebound things

[Nif A @niferential]

im seeing a lot of interesting use cases for jev but most people are using it wrong. stop treating it like a faster cheaper llm call.

what you should actually be using it for
- fast classification & routing at scale
- tool selection, retry/stop, risk scoring
- guardrail layer between your llm and the user
- basically any decision where the answers are known upfront and you need a calibrated pick

[Nif A @niferential]

@thdxr what about bots that sound like a human

[Nif A @niferential]

@thdxr Go pro

[Nif A @niferential]

release day an hour before doors close

[Nif A @niferential]

@thdxr local is the best if you want cheap tokens and want to maintain security

[Nif A @niferential]

@Trace_Cohen @Muse @Chase @Plaid @Adobe as much as I appreciate tools like muse and instinct a basic budgeting app can do this and some even track down and cancel unnecessary subscriptions

[Ryan Wright @IQReactorAI]

Guys and Girl nerds, the tiny 500m-1.2B models trained for a single task are taking off cause they are dirt cheap to run and do that task better then frontier models.

Us Local AI folks can run AND train these at home. (I will soon)

Actually ☝️🤓 For a training a highly specialized 1B model, 1,000 to 5,000 high-quality, varied examples are often enough to achieve frontier-level performance on that specific task. Not 50,000 to 1 million like with big models.

Checkout this blog article by @ liquidai where they talk about their tiny task models.

"LFM2-Extract – A 350M and 1.2B multilingual models for data extraction from unstructured text, like turning invoice emails into JSON objects.

LFM2‑350M‑ENJP‑MT –  A 350M model for bidirectional English ↔ Japanese translation.

LFM2‑1.2B‑RAG – A 1.2B model optimized for long‑context question answering in RAG pipelines.

LFM2‑1.2B‑Tool – A 1.2B model built for function calling and agentic tool use.

LFM2‑350M‑Math –  A 350M reasoning model for solving mathematical problems.

Luth-LFM2 – An additional community-driven series of French fine-tunes to enable general-purpose assistants for on-device chat."

Read more here: https://t.co/GqsOAhUD3C

[Nif A @niferential]

@omooretweets there were probably enough people asking variations of this question on google search way before llms got popular

[Nif A @niferential]

@colebemis The options are endless - right now using it to track different agent sessions

[Nif A @niferential]

@nizzyabi they got tired of blaming claude for breaking prod

[Nif A @niferential]

adding another screen to my workflow