Srinivas Njay

@andrewchen Check out interface.ai. We are building BankGPT.

Who’s working on this idea: Openclaw for personal finance - integrates w all your banks/cards/etc - understands tax returns and filings - monitors portfolio and competitors - digests proprietary data sources (credit card panels, app rankings, and etc) - reads company news and X Etc etc

Rep. Ro Khanna (D-CA) on Epstein Files youtu.be/5mKlnSp_FSM?si… via @YouTube @RoKhanna thank you!

If you are truly, really committed to excellence. I mean deeply. Many people will not really want to work with you. Push on

Interface.ai: Srinivas Njay raises $30M from Avataar Venture Partners to build customer automation platform for banks. @interfaceAI @srinivasnjay @AvataarVC

🚀 Exciting news! @interfaceAI just secured $30M to transform customer service in banking with AI. Discover more: tinyurl.com/4ca3dbxs @CsharpCorner @srinivasnjay #AI #Fintech #CustomerService #Banking #StartupNews #Innovation

1/ Can Large Language Models (LLMs) truly reason? Or are they just sophisticated pattern matchers? In our latest preprint, we explore this key question through a large-scale study of both open-source like Llama, Phi, Gemma, and Mistral and leading closed models, including the recent OpenAI GPT-4o and o1-series. arxiv.org/pdf/2410.05229 Work done with @i_mirzadeh, @KeivanAlizadeh2, Hooman Shahrokhi, Samy Bengio, @OncelTuzel. #LLM #Reasoning #Mathematics #AGI #Research #Apple

@arnavsahu341 In B2C, a poor product can spell defeat. In B2B, a bad sponsor choice can ruin the fleet.

I still cannot believe that Servicenow - with software that looks so old - has close to $10B in revenue and a $130B market cap. I understand - its annoying to move over to new software etc etc - but seriously - how are companies paying millions for this? Its basic UI stuff that can be fixed in 3 days.

Srinivas Njay, CEO and cofounder of interface. ai, has signed up more than 100 credit unions and community banks, with 16 million customers, to use his AI-powered virtual assistant. Read more: trib.al/cI2W9qV

@tim_cook Seems like Apple Vision Pro's preorder page missed a class in Dynamic Slot Allocation 101. Customers playing musical chairs with pickup times is one way to build anticipation... or a queue management logic is in desperate need of a debug session! 🤓 #TechFails #AppleVisionPro

Apple Vision Pro pre-orders begin today! We’re so excited for you to experience spatial computing for the first time!  apple.co/applevisionpro

Seems like Apple Vision Pro's preorder page missed a class in Dynamic Slot Allocation 101. Customers playing musical chairs with pickup times is one way to build anticipation... or a queue management logic in desperate need of a debug session! 🤓 #TechFails #AppleVisionPro

I touched on the idea of sleeper agent LLMs at the end of my recent video, as a likely major security challenge for LLMs (perhaps more devious than prompt injection). The concern I described is that an attacker might be able to craft special kind of text (e.g. with a trigger phrase), put it up somewhere on the internet, so that when it later gets pick up and trained on, it poisons the base model in specific, narrow settings (e.g. when it sees that trigger phrase) to carry out actions in some controllable manner (e.g. jailbreak, or data exfiltration). Perhaps the attack might not even look like readable text - it could be obfuscated in weird UTF-8 characters, byte64 encodings, or carefully perturbed images, making it very hard to detect by simply inspecting data. One could imagine computer security equivalents of zero-day vulnerability markets, selling these trigger phrases. To my knowledge the above attack hasn't been convincingly demonstrated yet. This paper studies a similar (slightly weaker?) setting, showing that given some (potentially poisoned) model, you can't "make it safe" just by applying the current/standard safety finetuning. The model doesn't learn to become safe across the board and can continue to misbehave in narrow ways that potentially only the attacker knows how to exploit. Here, the attack hides in the model weights instead of hiding in some data, so the more direct attack here looks like someone releasing a (secretly poisoned) open weights model, which others pick up, finetune and deploy, only to become secretly vulnerable. Well-worth studying directions in LLM security and expecting a lot more to follow.

Thank you for sharing. Another one of my favorite homestyle, authentic Karnataka foods in the Bay Area is @suggioota's Vidhaythri Bhavan-style masala dosa and Bangalore street food like masala puri, ragi mudde, and joladha rotti engai. These popular Karnataka dishes, which you don’t often find in any South Indian restaurant, are typically part of the menu and are very authentic.

If you are a South Indian living in the Bay Area you are pretty much taken care of in terms of food of w/ @DoorDash Shastha Food and @MylaporeExpress Doordash for restaurant food and groceries Shastha Food for batter and snacks Mylapore Express for daily vegetarian meals for the family @MylaporeExpress should be a case study in itself. Picked up during the pandemic and delivers food daily from Sacramento! It has to be quite a profitable business given the volume.

@jimmydouglas Experience is the best teacher, and only those schooled by it can truly be a master instructors. Kudos to you for stepping into the other side!

I made 25 angel investments before jumping off to take something from zero to one for the first time. To every founder I spoke to like I knew what it was like, I did not, and I'm sorry I was confidently wrong. This is so fun but impossible to understand without having done it.

Experience is the best teacher, and only those schooled by it can truly be a master instructors

🚀 Exciting times in AI! Here’s a quick roundup of some groundbreaking recent developments (please add any other notable developments in comments): ✨ **Orca-2 Over LLaMA-2**: Microsoft's Orca-2 has made waves by fine-tuning on LLaMA-2, showcasing reasoning akin to GPT-3.5. With a 13B parameter version in tow, Orca-2 imitates complex explanations from GPT-4, pointing towards smaller models achieving big results. Who said size is everything? 🏎️ **CMM's Speed Revolution**: Conditional Matrix Multiplication (CMM) just got a turbo boost! Researchers have hit a remarkable 78x speedup in feedforward network inference, with predictions of a staggering 300x in hardware code. This means faster, more cost-effective language model usage is on the horizon. 💡 **S-LoRA Algorithm**: Imagine running thousands of language models on a single GPU. Thanks to the S-LoRA algorithm, this could become a reality with up to 4x better performance. Fine-tuning large language models is about to get a lot more efficient and cost-effective. 🔮 **The Mystery of Q***: OpenAI's rumored Q* is shrouded in mystery, but it's speculated to be a leap towards enhanced AI reasoning. Could this be the dawn of true AI agents? Stay tuned as we unravel these AI advancements. The future is here, and it's running at 300x speed! 🌐 #AILanguageModels #TechInnovation #Orca2 #CMM #SLORA #AICommunity

🇺🇲/acc 🤝 e/acc 🚀 Watch the whole thing. Will you answer the call to build? 🫡

State of play: - Negotiations for @Sama’s return have hit snag over makeup of board - @satyanadella leading the charge on discussions - Board wants input on new board members before they walk away so they are going through potential new candidates one by one. @btaylor seems to have been cleared by all parties. - situation is *super dynamic* & could change any minute - Sam keeping his options open ie. returning or starting newco With @EdLudlow @dinabass @rachelmetz bloomberg.com/news/articles/…

Wow. I totally disagree also. Here's why: x.com/bgurley/status… More importantly, if he is right, and open source will never catch up (& only fall further behind), why use your lobbying efforts to block it? Why not let it become irrelevant in the market as you suggest?

Breaking: OpenAI board in discussions with Sam Altman to return as CEO trib.al/DTxl5tC