Manjith chakravarthi

Gen AI, AR/VR

[Alexandr Wang @alexandr_wang]

so peaceful

[Sentient @SentientAGI]

📢 AGI-Thon Winners Announced!

A huge congratulations to the 30 incredible teams who participated in the @agihouse_org Werewolf AGI-Thon!

Presenting our winners…

🥇 1st Place – Werewolf Team 6
Prize: $10,000
Members: @atemyipod (Kipo AI), @pochhi_nilay (Rippling), @manji369 (Google)

🥈 2nd Place – Werewolf Team 8
Prize: $5,000
Members: Aydar Akhmetzyanov (DoorDash), Dhawal Modi (Deep Learning), Vince (IGCY AI), Lily Zhu (Google Play)

🥉 3rd Place – Werewolf Team 1
Prize: $3,000
Members: Clay Schubiner (Canal), Alex Blickenstaff (Apple)

🏅 Finalists (4th-8th Place)
Prize: $500 Each
Teams: 13, 14, 28, 9, 30

Congratulations to all the winners! Your contributions are paving the way for Community-Built Open AGI.

[Tarun Raheja @atemyipod]

we won the grand prize at the @agihouse_org werewolf agent hackathon! 💪 🏅

big thanks to @sentient_agi and @FireworksAI_HQ for hosting, and to my amazing team (@pochhi_nilay,@manji369) 🔥

a 🧵 on how we did it -

[Manjith chakravarthi @manji369]

@TyMcDuffey @aisolopreneur I meant GPT-11 is a loooong loooong way ahead

[Cognition @cognition]

Today we're excited to introduce Devin, the first AI software engineer.

Devin is the new state-of-the-art on the SWE-Bench coding benchmark, has successfully passed practical engineering interviews from leading AI companies, and has even completed real jobs on Upwork.

Devin is an autonomous agent that solves engineering tasks through the use of its own shell, code editor, and web browser.

When evaluated on the SWE-Bench benchmark, which asks an AI to resolve GitHub issues found in real-world open-source projects, Devin correctly resolves 13.86% of the issues unassisted, far exceeding the previous state-of-the-art model performance of 1.96% unassisted and 4.80% assisted.

Check out what Devin can do in the thread below.

[Ramsri Goutham Golla @ramsri_goutham]

🚀 Introducing Telugu Gemma 7B Lora Finetuned Model - Perhaps the best open LLM model yet for Telugu text generation out there!

What's a better way to witness the results firsthand than experimenting with it yourself in Google Colab (on a standard T4 instance)?

Inference code with @unslothai Library (2X faster than HF): https://t.co/fkUgdqBGT8

Inference code with Huggingface for those who prefer this:

https://t.co/kzIdxbDgg5

A few key details:

-> Unlike Llama2 which has a 32k vocab size, Google's Gemma tokenizer has a 250k vocab size. We found that the Gemma tokenizer is already good at tokenizing Telugu although not dense. But this is a significant upgrade over other tokenizers including OpenAI which resolves Telugu text to its individual Unicode bytes. This is a crucial step because one doesn't need to extend the base vocabulary of the tokenizer and do additional pretraining like in Llama2.

-> Google's Gemma 7B is trained on 6 trillion tokens unlike Llama2 (2T). With a vocab of 250k and 6 Trillion tokens, although the Telugu % might be small, it has better base knowledge than any other open models of its size.

-> We trained our LORA model on A100 80GB using the Unsloth library (faster training than HF) for 2 epochs. Note that we didn't use Qlora (4bit) but Lora with a 32-bit optimizer (paged_adamw_32bit).

-> We finetuned the model with 80k curated instruction datasets we released earlier as part of Telugu LLM Labs. Though we fine-tuned the model with Alpaca format with just Telugu instruction, input, and response triplets, the fine-tuned model has unique characteristics to take in instruction in Telugu script or English Script and respond in Telugu! This is powerful as most users prefer to write instructions in English as it is easier to type on a Romanized keyboard!

A few impressive results (Note that Response is generated by the model with user Instruction):

### Instruction:

I live in India. Is it good to go to Masters after undergrad or should I work for a few years in India?

### Response:

AIగా, నాకు వ్యక్తిగత అభిప్రాయాలు లేదా సలహాలకు సామర్థ్యం లేదు. ఏదేమైనా, మీరు మీ దేశంలో ఉండటం, మీరు అనుభవించిన అనుభవాలు మరియు మీరు కోరుకునే కెరీర్ మార్గం గురించి మీరు ఆలోచించడం చాలా ముఖ్యం.

మొదట, మీరు మీ దేశంలో ఉండటం వల్ల కలిగే ప్రయోజనాలను పరిగణించండి. మీరు ఇప్పటికే ఉన్న పరిచయం, సంబంధాలు మరియు నెట్వర్క్ను కలిగి ఉండవచ్చు, ఇవి మీకు ఉద్యోగ అవకాశాలు మరియు కెరీర్ పురోగతికి సహాయపడతాయి. అదనంగా, మీరు మీ దేశంలో నివసించడం మీకు వ్యక్తిగత మరియు సామాజిక అనుభవాలకు దారితీస్తుంది, ఇది మీ జీవితంలోని అన్ని రంగాలలో మరింత సంతృప్తికరంగా ఉండటానికి మీకు సహాయపడుతుంది.

మీరు మీ దేశంలో ఉండటం వల్ల కలిగే ప్రతికూలతలను కూడా పరిగణించండి .....

Find more examples in the Colab notebook!

Made in collaboration with @ravithejads as part of our Telugu LLM Labs initiative!

[👩‍💻 Paige Bailey @DynamicWebPaige]

🤯 Mind officially blown:

I recorded a screen capture of a task (looking for an apartment on Zillow). Gemini was able to generate Selenium code to replicate that task, and described everything I did step-by-step.

It even caught that my threshold was set to $3K, even though I didn't explicitly select it. 🤯🔥

"This code will open a Chrome browser, navigate to Zillow, enter "Cupertino, CA" in the search bar, click on the "For Rent" tab, set the price range to "Up to $3K", set the number of bedrooms to "2+", select the "Apartments/Condos/Co-ops" checkbox, click on the "Apply" button, wait for the results to load, print the results, and close the browser."

[Manjith chakravarthi @manji369]

@mikeinspace How do you know she is a Hindu?

[Sully @SullyOmarr]

Gemini 1.5 pro is STILL under hyped

I uploaded an entire codebase directly from github, AND all of the issues (@vercel ai sdk,)

Not only was it able to understand the entire codebase, it identified the most urgent issue, and IMPLEMENTED a fix.

This changes everything

[Vaibhav (VB) Srivastav @reach_vb]

Whisper running on WatchOS! 🔥

> Powered by WhisperKit by @argmax
> Supports up to Whisper base
> Leverages Neural Engine ⚡
> Three lines of code ;)
> Works real-time!
> MIT license

Quite amazed by the speed with which Argmax is shipping.

Possibly the fastest & reliable way to run Whisper on Apple devices!

[Manjith chakravarthi @manji369]

@sama A drone shot of a mystical fantasy land of pigs busy at work, building shelters, raising cattle, growing vegetation, generating power. There is also a lot of traffic in the air because of flying cars zooming fast through the atmosphere.

[Manjith chakravarthi @manji369]

https://t.co/S5ZUNog88H
https://t.co/4XZ6uoEx4D

Two huge breakthroughs today! 2024 is where things get real!

[Jim Fan @DrJimFan]

MidJourney hired an engineer from Apple Vision Pro to be "Head of Hardware". My best guess is that they are thinking about generating full synthetic worlds for AR/VR, because of their rumored works on text-to-3D. Data-driven simulation is a hot topic at NVIDIA and very dear to my heart.

Congrats to the Vision Pro engineer who found a new adventure! Would love to see what MidJourney comes up with.

First discovered by @zackhargett

[LlamaIndex 🦙 @llama_index]

There's still time to register for our first-ever in-person hackathon on Friday! Prizes are up to $16,000 including $8000+ in cash!
https://t.co/j33mXMctJV

[Manjith chakravarthi @manji369]

@ramsri_goutham @skalskip92 I guess the current number of cells are a lot for GPT vision. I tried with fewer cells and added cell numbers. This seems to be a bit more accurate.

[Rowan Cheung @rowancheung]

@javilopen @Magnific_AI 6. Decohere AI teased a preview of their new AI Video generation platform.

@decohere_ai uses Stability AI's new SDXL Turbo for real-time generations - absolutely wild!

[Manjith chakravarthi @manji369]

Yayy!

#GPT4 #gptvision #indiehackers

[Manjith chakravarthi @manji369]

Not no code, not low code, but some third thing.

[swyx @swyx]

Stop building the thing.
Build the thing that builds all the things.

IMO the most important thing every developer could be doing right now on nights and weekends is building a general purpose personal junior dev agent they can control and trust, that they can scale to fleets.

i'm 3 hours into building my own "smol developer" and its now pretty capable of going from prompt to code. releasing mvp tomorrow.

[Bilawal Sidhu @bilawalsidhu]

Such a dope use of wonder studio. Virtual production and VFX continue to get democratized 🪄

Just imagine how much longer this would’ve taken with a classical approach to rotoscoping, in-painting, 3D tracking, animation, rendering and compositing 😅