Ankit Tripathi

Something has changed completely. I haven't slept for more than 4 hours in last 48 hours. > i opened terminal > ran llama. cpp > installed qwen 3.5 4B Q4 locally > installed qwen 3.5 9B Q4 locally > started testing side by side It made me shiver to the core! These models are super good at reasoning and instruction following. - great at logic - brilliant thinking pattern - super fast latency It forced me to download their base models (weights) and fine tune these models for specific use cases. I am currently performing distillation on Qwen 3.5 4B Q4 base and it surpassed GPT oss120B in reasoning, instruction following and tool calls. A tiny model beating 30x bigger model is no joke. Qwen 3.5 9B model is better than GPT-4o what is an OG model. What's interesting is these models doesn't require a large dataset to be fine tuned, they have amazing KV cache, and requires just 8GB to 12GB RAM to functional properly. I also downloaded the 4B model on my phone and it gave me 15 tokens/sec latency which is really good. I can take it to 25 tokens/sec. Local ChatGPT running on my phone. In 2 weeks I'll share my fine-tuned qwen model with you and I'll share how easy it is to: > prepare dataset using Ralph loop > distill models using Codex > to quantize a model without lossing performance. > deploying the models on consumer hardware (no fluffy ollama) These are exciting times.

@akshay_pachaar but building the high qualit data is still 80% job of fine tunning right?

HuggingFace just made fine-tuning 10x easier! One line of English to fine-tune any open-source LLM. They released a new "skill" you can plug into Claude or any coding agent. It doesn't just write training scripts, but actually submits jobs to cloud GPUs, monitors progress, and pushes finished models to the Hub. Here's how it works: You say something like: "Fine-tune Qwen3-0.6B on the open-r1/codeforces-cots dataset" And Claude will: ↳ Validate your dataset format ↳ Select appropriate GPU hardware ↳ Submit the job to Hugging Face Jobs ↳ Monitor training progress ↳ Push the finished model to the Hub The model trains on Hugging Face GPUs while you do other things. When it's done, your fine-tuned model appears on the Hub, ready to use. This isn't a toy demo. The skill supports production training methods: SFT, DPO, and GRPO. You can train models from 0.5B to 70B parameters, convert them to GGUF for local deployment, and run multi-stage pipelines. A full training run on a small model costs only about $0.30. Link to the full tutorial in the next tweet!

ok, so I built a claude code agent that find domains under $20 and buy them for me. > running autonomously > trying dozens of combinations > researching the web forums > using open source ai models only > costing $2/day to run what a time to be a domain nerd.

@_avichawla Hey avi nice explanation with diagramm quick question are you selling fine tunning llm services since you are doing it fro 2years or doing something different ? i am currently learning fine tunning llm and slm Drop some advice if you have !

I have been fine-tuning LLMs for over 2 years now! Here are the top 5 LLM fine-tuning techniques, explained with visuals: First of all, what's so different about LLM finetuning? Traditional fine‑tuning is impractical for LLMs (billions of params; 100s GB). Since this kind of compute isn't accessible to everyone, parameter-efficient finetuning (PEFT) came into existence. Before we go into details of each technique, here's some background that will help you better understand these techniques: LLM weights are matrices of numbers adjusted during finetuning. Most PEFT techniques involve finding a lower-rank adaptation of these matrices, a smaller-dimensional matrix that can still represent the information stored in the original. Now with a basic understanding of the rank of a matrix, we're in a good position to understand the different finetuning techniques. (refer to the image below for a visual explanation of each technique) 1) LoRA - Add two low-rank trainable matrices, A and B, alongside weight matrices. - Instead of fine-tuning W, adjust the updates in these low-rank matrices. Even for the largest of LLMs, LoRA matrices take up a few MBs of memory. 2) LoRA-FA While LoRA significantly decreases the total trainable parameters, it requires substantial activation memory to update the low-rank weights. LoRA-FA (FA stands for Frozen-A) freezes matrix A and only updates matrix B. 3) VeRA - In LoRA, low-rank matrices A and B are unique for each layer. - In VeRA, A and B are frozen, random, and shared across all layers. - Instead, it learns layer-specific scaling VECTORS (b and d) instead. 4) Delta-LoRA - It tunes the matrix W as well, but not in the traditional way. - Here, the difference (or delta) between the product of matrices A and B in two consecutive training steps is added to W. 5) LoRA+ - In LoRA, both matrices A and B are updated with the same learning rate. - Authors of LoRA+ found that setting a higher learning rate for matrix B results in better convergence. ____ Find me → @_avichawla Every day, I share tutorials and insights on DS, ML, LLMs, and RAGs.

See this....

@freddiebambo Outbound

GIRLFRIEND: "you're a professional texter making $40k/month" she's not wrong most businesses waste: - $10k/month on ads - $60k/year on sales reps - $5k/month on agencies linkedin has 60 million ceos you can DM directly problem: personalized messages take 15 min each my AI does it in 3 seconds: - finds prospects - researches them - writes personalized DM - sends 1,000+/month automatically 31% reply rate vs 3% industry 20-40 calls/month $700k in 2 years giving away: → video walkthrough → AI prompts → message templates → automation setup comment "OUTBOUND" must be following she still thinks i just text 😂

@AndrewKim @bubble Send some work towards me😅

Is it just me, or is this time of the year the busiest for @bubble devs? It's 3 straight years where I'm bombarded with work around November-January.

the AI Syndicate group chat is accepting new applications like and reply if you wanna be added [image generated by a gc member]

Ok here's the new flow: UI → Gemini 3.0 pro Code → Claude Sonnet 4.5 Plan → GPT 5.1 Just try this; remove all others.

Met a guy in Latvia who clears $10M/yr Made me look poor bc we met in his restaurant Legit psychopath. No emotion on his face ever. Drives a matte black Maybach. Makes decisions like a fucking robot He taught me something that changed everything about how I operate We were having coffee and I told him I lost $30k on a failed launch. I was pissed about it He looked at me dead in the eyes and said: "You're angry you lost $10k? You're a goon. The $10k is gone. The money is dead. But the data from the loss? That's the asset. Stop crying. Calculate the data" (translated, mf speaks russian) Then he went back to his espresso like nothing happened That hit me different You mfs are too emotional. You're "sad" about a failed launch. You're "anxious" about a sales call. You're "stressed" about the algorithm Meanwhile killers feel nothing. They just execute. They calculate. They move Lost another $15k on a campaign 2 weeks ago Old me would've been depressed for weeks. Questioning everything. Probably would've given up New me analyzed the data in 4 hours. Found the 3 variables that killed it. Fixed them. Your emotions are making you easy to kill in business Become a psychopath or stay broke study the yegor method

everyone's using Perplexity wrong and wondering why their business isn't printing money the tool is right there giving you perfect market intelligence and you're treating it like google with a chat interface. here's what broke info product creators do: they ask Perplexity surface-level questions about their niche, get generic answers, maybe compile them into a notion doc, never actually use the intelligence to build anything that prints. instead you can use Perplexity as a competitive intelligence weapon that tells you exactly what to build and who to sell it to here's the specific research process that generated $180k in the last 6 months: step 1: map the entire pain landscape of your target market using search clusters i don't ask "what problems do crypto traders have" i run 15-20 highly specific queries designed to expose the exact language, price points, and failure modes of my target buyers queries like: "what are crypto trading firms complaining about on reddit in 2025" + "what automation tools do $10M+ trading desks actually use" + "why do crypto traders fire their operations people" each query engineered to extract signal that competitors miss. step 2: reverse engineer every successful competitor's positioning using their own content against them here's what this looks like: "analyze the last 6 months of [competitor]'s twitter content and identify their core value propositions" + "what gaps exist in [competitor]'s service offering based on customer complaints" Perplexity pulls real data from real conversations. you now know exactly what they're selling, what they're missing, and where to position against them. step 3: validate pricing with zero guesswork normal people guess at pricing or worse, they ask their audience "what would you pay for this?" that's how you end up with $47 products serving broke people. instead: "what do enterprise clients pay for [similar service] in [industry]" + "what's the typical retainer structure for [your service type] serving $100M+ companies" Perplexity aggregates real deal data from actual contracts, case studies, and service agreements. i discovered my initial $8k/month pricing was leaving $15k on the table per client just by running 3 targeted queries about competitor pricing in the web3 infrastructure space. step 4: steal every successful hook, angle, and framework from adjacent markets. this is where people fuck up completely. they stay in their echo chamber researching their own market. real intelligence comes from adjacent markets solving similar problems differently. query structure: "what frameworks do [adjacent industry] use to solve [similar problem]" + "best performing content angles in [adjacent niche] for [similar outcome]" example: i'm selling AI automation to trading firms, so i research how devops consultancies sell infrastructure to fintech, how trading educators sell systems to retail traders, how cybersecurity firms sell to regulated industries Perplexity finds the proven angles, i adapt them to my market, everyone thinks i'm a genius when i'm just stealing intelligently step 5: build your entire content strategy from actual search demand, not vibes here's the query that changed everything: "what questions are [target market] asking on reddit, and forums about [your topic] that have no good answers" Perplexity returns a list of underserved questions with real demand. you now have 6 months of content that's guaranteed to resonate because you're answering questions people are actively searching for, not questions you think they should care about here's what this produces in practice: before using this research process: creating content based on gut feeling, getting 200-400 impressions per post, zero inbound leads, spending 40 hours/month on content that doesn't convert after implementing systematic Perplexity research: every piece of content engineered from validated demand signals, average 3,000-8,000 impressions per post, 2-3 qualified inbound leads per week, spending 8 hours/month on content that actually prints content effectiveness increased 15x inbound lead flow went from zero to 8-12 qualified conversations monthly. close rate on those inbound leads: 67% because they're pre-sold by content that speaks their exact language here's why this works: you're no longer guessing about market demand, you're extracting intelligence from millions of real conversations and search patterns every competitor is creating content from their perspective. you're creating content from aggregated market intelligence about what actually works the gap between those two approaches is the difference between $3k/month and $150k/month here's the critical mistake most people make with Perplexity: they ask it questions like they're talking to a person. wrong. you're not having a conversation, you're running intelligence operations. every query should be designed to extract specific, actionable data that informs a business decision "what's the best way to do X" is a conversation question "what exact language do $10M+ companies use when describing problem X in their job postings and RFPs" is an intelligence query one gives you generic advice, the other gives you the exact words that make enterprise buyers lean forward here's what separates people who build 6-figure info businesses from people who stay stuck: the winners treat research as competitive intelligence extraction, the losers treat it as learning you don't need to learn more about your market, you need to extract decision-grade intelligence that tells you exactly what to build, how to price it, who to sell it to, and what words make them buy Perplexity gives you all of that if you know how to ask people don't know how to ask because they're scared to be specific about what they actually want to build they ask vague questions and get vague answers and wonder why their info business never launches people who win run surgical queries designed to extract the exact intelligence needed to make the next decision that's the difference.

in last 120 days, chinese models released: > kimi-k2-thinking (1t-a32b) > minimax m2 > deepseek v3.2 > glm-4.6 (335b-a32b) > qwen3-vl-30b-a3b (thinking) > qwen3-vl-235b-a22b (thinking) > qwen3-next 80b-a3b (thinking) > glm-4.5v (vlm, 106b-a12b) > deepseek v3.1 > doubao 1.6-vision (multimodal) > doubao translation 1.5 (28 languages) > ernie x1.1 (reasoning) > hunyuan-mt-7b & chimera-7b > minicpm-v 4.5 (8b) > internvl 3.5 (multimodal 1b to 241b) > step-3 (vlm, 321b/38b) > sensenova v6.5 (sensetime, multimodal) > glm-4.5 air (base & instruct) > glm-4.5 (base & instruct) > qwen 3-coder-30b-a3b (thinking) > qwen3-coder-480b-a35b (thinking) > qwen3-30b-a3b-2507 (thinking) > qwen3-235b-a22b-2507 (thinking) >* kimi k2 (1t-a32b) these are all open source models. download, deploy on prem, or on private cloud. imagine 1 year from today. open source ai will surpass closed source ai in many fields.

> be chinese > don't cry about GPUs > build best LLM models > make them open-source > drop them on X with tiny tweet > get back to work =/ don't be west!

the problem with the AI space is that most of these gurus and influencers just yap horseshit i personally know at least 20 other AI agency founders and most of them suck ass at tech 90% of them have never touched code in their life yet they're out here teaching others how to build AI agents "here's the n8n workflow" doesn't sound good when coming from a person who doesn't know shit about no-code tools you learn what a "node" is and start selling courses like you've already made a million dollars but all of this is going to change soon, i'm bringing something new and massive i've personally spent YEARS coding as a dev, using no-code automation tools and even signed 6-figure deals with the largest tech giants in the space i can't even take their names on here because that'll disrupt your mind, but you must know that i could do all that only because i actually spent years learning things and applying them i can DESTROY any dev on here if i want to but i'm too busy making money working with billion dollar organizations soon i'll be revealing how i did it and how you can do the same until then enjoy the show :)

@bindureddy Correct me if I am wrong are they working on building slm or alternative of llm but with less build and computer cost

@_ProfitPath yeah i am suffering with this iam over palnning daily ,doin research,worried that my idea will not work and i will not make money..what i should do

your "perfect" business plan rotting in google drive while some 90 IQ dropout makes $50k/month with a 3-page PDF... the difference? he has psychotic confidence and you have imposter syndrome. i watch this shit every day: smart people with brilliant ideas paralyzed by self-doubt, "researching" for the 90th month, waiting for the "right time" meanwhile complete retards with objectively terrible products closing $5k deals daily because they pitch like they're selling oxygen to drowning people. this kid i know can't spell "entrepreneur" but he DMs 100 people daily with the energy of someone handing out million dollar checks. his funnel? broken. his copy? chatgpt garbage. his product? screenshots from reddit. but when he gets on a call he talks about his offer like it's the second coming of Christ... no hesitation, no "um maybe if you want", just pure delusional certainty that what he has will change your life. and people buy it. not because it's good... because conviction is contagious. you're sitting there with your MBA and your market research and your competitive analysis wondering if you're "ready" he's out here with a room temperature IQ and the unshakeable belief that he's god's gift to business. the market doesn't give a fuck about your credentials or your perfect plan or your imposter syndrome... it buys from whoever believes hardest. your brilliant idea presented with uncertainty = $0 his dogshit idea presented with psychotic confidence = $600k/year stop "validating", stop "researching", stop "waiting"... start executing with the delusional confidence of someone who can't comprehend failure. the world rewards psychotic conviction over competent hesitation every single time. your perfectionism is poverty. his delusion is wealth. pick one.

@imtommitchell What you think about data analyst role...is it ai proof or only senior role will be available

Recruiters are your best friend when it comes to landing data roles. Here's a hack to find them quickly on LinkedIn. Copy and paste this into the search bar: "recruiter" AND "data analysis" Or whatever role you're looking to land. Filter by location and click people tab. Reach out, be friendly and connect. Don't be pushy - make a genuine attempt to network. You never know what it could lead to.

over the next few days i'll be dropping some sauce about how you can start building, systemizing and integrating AI infrastructures to charge $10k-20k per project i'm going to cover the following topics: - how to pitch in UHNW industries (i personally know some people who sell $100k+ worth of automations here) - how to build your USP to stand out from every other AI bro selling commoditized automations - how to use code and no-code tools as a beginner and create infrastructures worth multi 5-figures - how to create the most outrageous automations and become a frontrunner in your industry/niche - how to stop relying on just n8n and learn how to code automations that actually sell - how to present yourself to prospects so they pay your invoice of $10k+ without thinking twice - how to position yourself as an asset, not just another ai bro selling agents and a lot more if you want me to cover anything else, drop in the comments below if not, stay tuned and enjoy the sauce :)

@paoloanzn audit

your are probably more skilled than 90% of AI gurus out here but yet you are still broke as fck... you know why? most devs build demos. you need to build revenue systems. i'm leaking the FULL architecture behind a Lead Audit Agent that's currently running in production and generating qualified leads worth $10K+ monthly for clients this ain't a "prompt engineering guide" (regurgitated by chagpt) this ain't a langchain tutorial this ain't some "HOW TO DO X with AI and make $$$" from someone who has no fucking idea of how to do that in the first place this is production-grade agent architecture that prints money if you're serious about seeing how money-making ai systems are actually built - from the orchestration logic to the business integration - this is it the first doc from the AI Ops Classified series follow, RT & comment "audit" and i'll send you the full system breakdown study how it's built. reverse engineer it build your own version that makes YOU money.