Sumit Gupta

@andrewdsouza @boardyai Boardy

I want to help 100 founders raise their rounds before the end of the year. We just unlocked a version of @boardyai that we didn’t think was possible. Boardy raised his own $8M seed round earlier this year. Since then, he's introduced thousands of founders to investors who have raised millions. We’ve been testing an overpowered version internally, and the results have been wild - founders are meeting stronger investors and closing rounds in weeks instead of months. He’s not ready for public release yet (the token costs are insane), but it’s time to start introducing him to the world. I’m personally working with 100 founders who are actively raising to help them close before the end of the year. You’ll work directly with me and the Boardy team and get early access before this version rolls out next year. Comment “Boardy” below, and I’ll message you the link to apply. If you know someone who’s raising, tag them below.

@rushabtated4 Please share

1/n Forget Fine-Tuning: Many-Shot Learning is the Future of AI Imagine a world where machines can learn complex tasks, not through laborious training on mountains of data, but by simply observing a few examples, much like humans do. This is the promise of In-Context Learning (ICL), a remarkable ability exhibited by large language models (LLMs) to adapt and acquire new skills on the fly. However, like a captivating story glimpsed through a keyhole, our view of ICL has been limited, restricted by the narrow window of context that LLMs could previously handle. Now, imagine flinging open that window, allowing a flood of information to illuminate the full potential of ICL. This is the exciting frontier explored in "Many-Shot In-Context Learning," a paper by Google that delves into the transformative power of scaling ICL to unprecedented levels. Prepare to be captivated as we journey beyond the limitations of the past and discover how many-shot learning can revolutionize the capabilities of LLMs, paving the way for a future where machines learn and adapt with human-like flexibility.

Gisting is a method proposed in a new paper to compress prompts into smaller representations called "gist tokens", in order to improve the efficiency of prompting large language models. Here is how gisting helps address the key pain points: Compression - By compressing full prompt instructions into just 1-10 gist tokens, gisting dramatically reduces the sequence length needed to encode prompts, freeing up space in the context window. Reuse - The gist token activations can be cached and reused across many queries, avoiding repeatedly encoding the full prompt. This saves compute. Generalization - Unlike finetuning/distillation which requires training each model per prompt, gisting learns to compress novel prompts at test time with no additional training. Attention masking - Gisting models are trained via a simple modification to Transformer attention masks that prevents attending from tokens after the gist to tokens before. This forces prompt compression at no extra cost. Efficiency gains - Experiments show gisting enables up to 26x compression rates, 40% FLOPs reduction, and 4.2% faster inference compared to unoptimized prompting. Gisting also enables caching prompts in encoder-decoder models. So in summary, by compressing prompts into reusable gist tokens, gisting improves efficiency and generalizes to new prompts, avoiding the drawbacks of both prompting and per-prompt finetuning.

@AidanLaBreche VIRAL !

62 prolific creators. 13,844,000 followers. I spent 6 months and 200+ hours breaking down 2,032 pieces of their best content in my "Viral Inspiration Lab". 𝗜𝘁'𝘀 𝗙𝗿𝗲𝗲 𝗳𝗼𝗿 𝘁𝗵𝗲 𝗻𝗲𝘅𝘁 𝟮𝟰𝗵𝗿𝘀. Like and comment VIRAL, I'll DM it to you. (Must be following!)

@olivercookson Please send

I bootstrapped Myprotein to a ≈ $500M exit (inc. secondary) That wouldn't be possible without: • Delegation • Strategic hires • Resource management I created a 14-page in-depth guide on the above. Like + Comment "Send" & I'll DM it to you for free (Must be following)

@Hamza_x8 @raoufmoulkaf17 land

inspiration leads to Creativity That’s why I spent +10 hours creating a figma file that contain most optimised landing pages for this week + explanations It’s for FREE for the next 24hours Like & Comment “land” And I’ll DM it to you for FREE (Must Follow)

@raoufmoulkaf17 Please share

🧲 50 high-converting landing pages Of 8-9 figures brands The ultimate collection (you won't find it anywhere) Take it for FREE now (24h) Just "like" and "retweet" And it's yours (most follow)

@Laraacostar Please send

LinkedIn has CRAZY organic reach. But most people take too long to write their posts. So I put together a quick guide to writing valuable LinkedIn content in 10 minutes or less. Want me to DM it to you for free? Follow me Like & Reply "Send" RT this tweet

@heyBarsee free

@INOXMovies So much for Covid protocols or basic hygiene practices that the straws provided at insignia is broken and drawn out of pockets and moneybags !

We had a spectacular and safe birthday celebration @INOXMovies Private Screening ! Thank you to the team at Inox,City Centre Kolkata,for being such courteous hosts.

@ashishmundhra11 @SonuSood @SoodFoundation @BloodDonorsIn @BloodAid @RedVolunteersWB @FindBlood @prateekpjjain Please look into this

Urgent Request: Bombay Oh+ve/Hh blood group. Location:Kolkata RGKMCH Hospital.Reqd Unit-1.Attender #9214027583.Verified 06/06/2021 #BloodMatters @SonuSood @SoodFoundation @BloodDonorsIn #rare @BloodAid @RedVolunteersWB @FindBlood #Critical #Help #SOSKolkata #bloodbank

#Help @Bloodmates1 @ThinkTinkAcad @ushsenush @nilormi_das @rishi7_roy @swastika24

#EverydayArmy #SimplyBlood @Simply_Blood #BloodMatters #Kolkata @iCanSaveLife @BLEED2SAVE @RaktHarPal @bloodconnectorg @BloodAid @Bloodhelpline #GiveBlood

@swatiroychowd12 @plf_savelives #life #Baby #Delivery Pls help

@TeamSOSIndia @TheRaktSewa #UrgentRequest

SOS request for B+ve platelets for a 12 year old boy ,currently at Apollo Banerghatta Road,Bangalore. (Posting on behalf of attender) Attendant Number- 7760414080,9980129040 @BloodDonorsIn @BloodAid @SonuSood @dhoondh @Lokalwala3 @dhoondh @TheRaktSewa @TeamSOSIndia @Hemkunt_Fdn