Owen Colegrove @ocolegro Twitter profile | Pikagi

Pikagi

Owen Colegrove

@ocolegro

5,255

Followers

668

Following

205

Media

2,075

Statuses

Physicist | Quant | Founder

San Francisco

https://t.co/D0V4Zo3WQb

Joined January 2021

Don't wanna be here? Send us removal request.

Pinned Tweet

@ocolegro

Owen Colegrove

3 months

Today SciPhi is open-sourcing Triplex, a SOTA LLM for knowledge graph construction. Triplex is so small that it can be used with SciPhi's R2R to build knowledge graphs directly from your laptop. Triplex outperforms few-shot prompted gpt-4o at 1/60th the inference cost.

17

58

536

Last Seen Profiles

@xng1012020

@VivianeSassi1

@stw_pdg

@kong_some_

@bokeplokalmalam

@binloji

@Masaemoooon

@morinatsu_LGBT

@BinorRaja

@MonavaleVlei

@hon_minato

@taesanarchives

@ouiytfdf7

@GovindHudda3

@PeteButtigieg

@Kimikii_x

@atommania

@omgshanti

@nallagal

@wentzeldbn

@eboybamie

@sa1991ii

@cyafuku

@bar_489

@RamKMadhvani

@SadieBugSummers

@DAINTYFAT

@nongmewy69

@preordercory

@2FOUND_2

@stiize

@sprinklesherron

@Barbiiie007

@futurehypehoops

@Gemma_D84

@ZenciT

@ocolegro

Owen Colegrove

10 months

Today, I'm sharing AgentSearch, an open-core effort to make humanity's knowledge accessible for LLM agents. To start, I have embedded all of Wikipedia, Arxiv, filtered common crawl, and more. The result is over 1 billion embedding vectors, try it at .

Tweet media one

61

256

2K

@ocolegro

Owen Colegrove

9 months

This result is so fascinating: WikiChat: Stopping LLM Hallucination - Achieves 97.9% factual accuracy in conversations with human users about recent topics, 55.0% better than GPT-4! Anyone interested in replicating + extending w/ AgentSearch?

Tweet media one

21

155

1K

@ocolegro

Owen Colegrove

1 year

We now have a pipeline that goes: MIT OCW -> Syllabus -> Table of Contents -> Textbook. The final step done w/ RAG over all of wikipedia. I see why Anthropic thinks they can auto research within in a few years. First book is here: []. More coming soon.

Tweet media one

Tweet media two

10

73

542

@ocolegro

Owen Colegrove

3 years

1.) As a builder on #ALGO , the recent price action does not worry me. In fact it excites me 👇

30

131

584

@ocolegro

Owen Colegrove

1 year

Our textbook generation is really cooking now. Now putting out ~500m tokens/day of RAG-grounded synthetic textbooks with a fine-tuned Mistral 7b. That's a good fraction of a public library. Thx in part to the help from @winglian

Tweet media one

24

43

535

@ocolegro

Owen Colegrove

1 year

If you are GPU-Poor, then you should probably be paying attention to small LLMs & synthetic data. Below is a list of the most interesting papers relating to this that I have found from the last year (ordered by release date), what am I missing? Self-Instruct

Tweet media one

10

66

354

@ocolegro

Owen Colegrove

1 year

With LLMs we can create a fully open-source Library of Alexandria. As a first attempt, I have generated 650,000 unique textbook samples from a diverse span of courses, kindergarten through graduate school. They are all available here: []

Tweet media one

16

41

318

@ocolegro

Owen Colegrove

1 year

We encountered issues with accuracy in synthetic data generation. Fortunately, a great approach called self-rag was introduced by @AkariAsai et al. Introducing SciPhi-Self-RAG-Mistral-7B-32k I fine-tuned the SciPhi model with this and more. The result is powerful + efficient.

Tweet media one

4

50

318

@ocolegro

Owen Colegrove

11 months

After the release of Phi-2, I felt it was a good time write-up my learnings from early efforts at replicating Phi-1.5. The number one insight is: Generating 20 billion unique, high-quality synthetic tokens is challenging. I explain why in depth here 👇

4

27

274

@ocolegro

Owen Colegrove

1 year

WizardCoder-34B appeared to be a big step forward in open-source AI models, but I find it falls far short of GPT-4! After reproducing their HumanEval and assessing on ~400 OOS LeetCode problem, I see that it is more on par w/ Claude-2 or GPT-3.5. See results in img below 👇

Tweet media one

17

21

168

@ocolegro

Owen Colegrove

9 months

Recently, engaging with developers during my time at YCW24 revealed a valuable insight: the demand for the advanced Retrieval Augmented Generation (RAG) solutions I envisioned is overshadowed by the need for simpler, more intuitive tooling. I'm grateful that YC's culture pushes

Tweet media one

20

16

181

@ocolegro

Owen Colegrove

8 months

I've been programming w/ Claude-3 Opus today, it feels like a step up from gpt-4. It is often providing better code and w/out abbreviation so that I can immediately copy + paste the output.

16

15

209

@ocolegro

Owen Colegrove

11 months

DeepSeek team doing some truly outstanding work. Their latest model is close to GPT-4 across a variety of high quality evals, and looks by far the best open option. It seems like there might be some advantage to training on both Chinese and English.

Tweet media one

10

27

206

@ocolegro

Owen Colegrove

4 months

I've been hacking with auto-generating KGs all week, and the tech is way more useful than I expected. It is cheap enough to run, and it gives LLMs the ability to handle a new class of population-level queries they can't yet, like "Find all the founders that worked at Google."

Tweet media one

Tweet media two

16

15

209

@ocolegro

Owen Colegrove

9 months

The full dataset for AgentSearch-V1 is now available on HF!! Recommended: @qdrant_engine - for indexing and search @nomic_ai - for visualization I'm looking to expand what is indexed - agent specific data like SEC filings, the Legal Pile, and whatever else you suggest.

@ocolegro

Owen Colegrove

10 months

Today, I'm sharing AgentSearch, an open-core effort to make humanity's knowledge accessible for LLM agents. To start, I have embedded all of Wikipedia, Arxiv, filtered common crawl, and more. The result is over 1 billion embedding vectors, try it at .

Tweet media one

61

256

2K

6

39

191

@ocolegro

Owen Colegrove

9 months

Today I hacked up an exciting upgrade to SciPhi's search. It is now powered by an improved model, and it also offers more comprehensive search results under the hood (Bing+AgentSearch). This rollout is combined with a very sleek new web app, thanks to @jiayq 's great work.

11

32

144

@ocolegro

Owen Colegrove

7 months

Ok twitter, lend me your strength - The day has finally come where we have a product truly worth putting in peoples hands. Please take for a spin (and show us some love on PH)

7

10

153

@ocolegro

Owen Colegrove

9 months

@arthurmensch "over-enthusiastic" That's a very polite way to describe their behavior.

1

0

178

@ocolegro

Owen Colegrove

7 months

I'm excited to finally put what we've been working on out there!

@ycombinator

Y Combinator

7 months

SciPhi (YC W24) is a cloud platform that simplifies the scaling of production-ready RAG systems. Built on the open-source R2R framework, SciPhi lets builders focus on creating innovative applications. Congrats on the launch, @ocolegro !

0

10

82

17

6

149

@ocolegro

Owen Colegrove

1 year

I've been spending some extra time working on self-coding agents. The toughest part? Crafting just the right amount of context to solve complex coding problems. This journey has culminated in a new search algorithm I've coined 'SymbolRank'. Ready to dive into the intricacies?

Tweet media one

Tweet media two

6

15

117

@ocolegro

Owen Colegrove

1 year

I've written a succinct write-up about this development and it's significance. The article is titled "AI Can Write Near Human-Level College Textbooks". In it, I explain how we discovered this realization and discuss its potential implications.

@ocolegro

Owen Colegrove

1 year

We now have a pipeline that goes: MIT OCW -> Syllabus -> Table of Contents -> Textbook. The final step done w/ RAG over all of wikipedia. I see why Anthropic thinks they can auto research within in a few years. First book is here: []. More coming soon.

Tweet media one

Tweet media two

10

73

542

7

16

129

@ocolegro

Owen Colegrove

10 months

Experimental Research Mode is Now Live at – Just Press 'Research'. The attached video illustrates research mode in action, and the algorithm is given below. If chatting with a document is the 'hello world' of LLMs, then consider this the 'merge-sort'."

8

17

147

@ocolegro

Owen Colegrove

1 year

Just pushed my first dataset to the hub []. ~300k Python coding samples created by WizardCoder-15B. They came out surprisingly good the process was very easy. I'm will extend to O(10M) samples and work on some related datasets. Feedback is welcome.

Tweet media one

@ocolegro

Owen Colegrove

1 year

I'm working on a configurable framework to streamline synthetic data generation & LLM cross-eval. It supports EvolInstruct and Textbooks w/ major LLMs []. I am using this to replicate Phi-1 and wanted to share since I think it will be helpful for others.

Tweet media one

2

13

89

4

22

133

@ocolegro

Owen Colegrove

8 months

I've been working on an open source framework, named R2R, to make it easier for developers to build, observe, and optimize RAG. It facilitates deployment and observability out of the box. We will be adding more features quickly.

13

11

67

@ocolegro

Owen Colegrove

3 years

1) In just three weeks the Algofi DEX will be on the $ALGO mainnet 🧵

1

25

133

@ocolegro

Owen Colegrove

1 year

We've had an ongoing discussion in Discord on optimizing the textbook generation pipeline. It quickly became clear that we need to rigorously measure the impact of RAG. We used Kaggle's recent competition as an inspiration to build a RAG Eval Harness. It's live in SciPhi now.

Tweet media one

3

13

103

@ocolegro

Owen Colegrove

10 months

Watch below as Sensei uses AgentSearch to do unsupervised research on 'Introduction to atomic physics'. I'm repeating this to do a depth-first search over 100k queries generated from a tree of knowledge spanning humanities academic understandings... this is getting interesting.

3

7

120

@ocolegro

Owen Colegrove

3 years

Who thinks @algofiorg should leverage State Proofs and the “NEAR trick” to build a trustless decentralized bridge to and from ETH?

11

12

113

@ocolegro

Owen Colegrove

1 year

Finally, we have a long term 8x A100, accelerating our open synthetic data research. First Colab Meeting: -=Title=- Powerful Models on a Budget: Combining DoReMi with Synthetic Data -=When=- Today @ 5:30 pm EST -=Why=- Discuss results & coordinate -=Where=- Discord

Tweet media one

2

4

66

@ocolegro

Owen Colegrove

6 months

Looks like Llama3 is not a MoE - bearish

26

3

113

@ocolegro

Owen Colegrove

3 years

13.) #Algofi will be coming along for the ride and innovating every step of the way. That is what I mean when I say "we are building the new financial system".

3

5

112

@ocolegro

Owen Colegrove

1 year

I'm working on a configurable framework to streamline synthetic data generation & LLM cross-eval. It supports EvolInstruct and Textbooks w/ major LLMs []. I am using this to replicate Phi-1 and wanted to share since I think it will be helpful for others.

Tweet media one

2

13

89

@ocolegro

Owen Colegrove

1 year

I have been grinding on a plug & play solution for OSS LLM + RAG over a large hosted knowledge engine (just wiki for now). You will be able to self-host the LLM, or access directly over an API. I've been working on a special fine tune to accompany the db. Any feature reqs?

Tweet media one

3

11

110

@ocolegro

Owen Colegrove

7 months

I have been obsessed with LLMs since ChatGPT came out. No experience since first getting cable internet has compared. I've experimented a lot with what to build and landed on a framework, R2R, to help devs build better RAG systems. Here's why I think it's so worthwhile -

3

4

94

@ocolegro

Owen Colegrove

11 months

I'm embedding a huge swathe of the web w/ SOTA (jina-v2-base). I will open source when I'm done. I'm using quality filtered RPJv2 as the first source. I generate titles before embedding in chunks [256,512, and 1024] and store in a sharded vector db. Any thoughts or requests?

Tweet media one

13

9

98

@ocolegro

Owen Colegrove

1 year

Phibrarian Alpha is live and ready for stress testing and user feedback. It's our first model checkpoint from a Mistral-7b fine-tune, harnessing over 1 billion tokens from open-phi and other sources, merging synthetic textbook data and more. The early results are promising.

Tweet media one

4

5

35

@ocolegro

Owen Colegrove

11 months

This Yi-34B model tops the leaderboard after HF's new, aptly named, DROP benchmark was added. Has anyone tried this model? Also... a lot of overfit models got crushed, but I'm sure people will just start overfitting this new eval soon enough.

Tweet media one

2

15

92

@ocolegro

Owen Colegrove

7 months

YC co's SciPhi uses for RAG cloud offering 1.) @lanterndb - Scaleable vector db for our users 2.) @hatchet_dev - Amazing workflow orchestration 3.) @reductoai - Custom trained LLM for pdf extraction 4.) @PareaAI - Online pipeline evaluation Who still says YC is just wrappers?

8

12

71

@ocolegro

Owen Colegrove

10 months

Serious question - can HF handle my 4tb dataset? Looking into pushing AgentSearch today.

15

2

88

@ocolegro

Owen Colegrove

1 year

Head to if you would like to test with the self-rag fine-tune. It's surprisingly good for educational context & I will be iterating to improve it over the coming days. Note, click the magnifying glass to see the fetched context. Now, I'm off to bed.

Tweet media one

6

9

81

@ocolegro

Owen Colegrove

1 year

So, I looking into embedding the most textbook/wiki like pages of the recent redpajamas drop, but even the top 0.2% of entries don't look so great when spot checked. How the F do LLMs manage to learn anything from these giant scrapes? Also, has anyone worked out a good filter?

Tweet media one

9

7

87

@ocolegro

Owen Colegrove

3 years

4.). #blockchain technology is here to stay in our world. A bear market is going to bring market participants back to reality and may act as a catalyst for accelerated technological growth.

2

6

80

@ocolegro

Owen Colegrove

1 year

@yacineMTB @snowclipsed Yes, best programmer I know is a former particle physicist. On a side note - some of the worst programmers I've met were also physicists :). Some (like myself once upon a time) think they are too smart for rigorous programming and instead pump out piles of spaghetti code

4

2

80

@ocolegro

Owen Colegrove

3 years

5.). Why is it a catalyst --> Users are going to feel a lot worse spending $50 on each transaction after seeing their portfolio go up in smoke.

3

5

81

@ocolegro

Owen Colegrove

3 years

12.) Algorand, or a chain just like it, will win the space. The amount of growth they will see in the next 20 years will be mind boggling - the worlds financial system is going to be transformed.

3

3

80

@ocolegro

Owen Colegrove

9 months

I'm thinking about going and writing an article for every requested article on Wikipedia and hosting the result. Any thoughts or ideas for other resources that contain similarly high signal requests for content that is worth autonomously authoring?

Tweet media one

Tweet media two

5

7

42

@ocolegro

Owen Colegrove

8 months

Shares fall 28% after Klarna says its AI assistant does the work of 700 full-time agents Is this the first time we've seen so many real world jobs disrupted by recent LLM advancements at publicly traded company?

8

7

77

@ocolegro

Owen Colegrove

3 years

11.) As the rotation intensifies users will become more informed. People will finally begin to appreciate the decentralized, scaleable, and and reliable properties of Algorand.

1

5

74

@ocolegro

Owen Colegrove

2 months

RAG requests serviced by R2R surpassed 5 million today. It's exciting to see so many developers accelerating their projects with this tool! The API is fairly stable now and we plan to keep iterating on all the great feedback to build the best end-to-end production RAG system.

Tweet media one

0

0

75

@ocolegro

Owen Colegrove

3 years

PSA: Be very weary of any token is a pre-product. $ALGO ecosystem is riddled with scammers, similar to ICO boom era of $ETH. They see an opportunity to take advantage of a new DeFi ecosystem. Real builders (C3, Tinyman, Algofi) focus on delivering value first, token second.

@AlgoFamily

MJ.algo

3 years

Facts regarding LOUDeFi: 1. Liquidity was pulled from $LOUD 457819394. 2. Team did not tell anyone before pulling liquidity. 3. Team is anonymous. 4. Team claim its not a #rugpull : Your thoughts? Legit or rug? Stay safe out there $algo family.

32

18

59

5

12

74

@ocolegro

Owen Colegrove

3 years

10.). People will begin to take note of the fact that #Algorand , #Solana , and a few others have designed robust solutions - their fees will stay low, they will scale.

6

3

72

@ocolegro

Owen Colegrove

1 year

Calling all Kaggle GMs. We’ve got to embed the entirety of human knowledge and then optimize a giga RAG system over that. This is how we will make LLMs truthful.

Tweet media one

5

3

71

@ocolegro

Owen Colegrove

1 year

I'm attempting to reproduce Phi-1 using only open-source data. My 320M parameter model has demonstrated the ability to solve some basic programming tasks, illustrated below. However, I still haven't been able to get my hands on an 8x A100 setup to scale. Recommendations?"

Tweet media one

Tweet media two

7

8

71

@ocolegro

Owen Colegrove

10 months

AFAIK, this represents the most extensive dataset of embeddings from recent SOTA. This work was enabled by innovative new initiatives like RedPajamas. The search engine employs a simple hybrid hierarchical approach, and is open sourced at .

Tweet card media

GitHub - SciPhi-AI/agent-search: AgentSearch is a framework for powering search agents and enabling...

AgentSearch is a framework for powering search agents and enabling customizable local search. - SciPhi-AI/agent-search

1

9

69

@ocolegro

Owen Colegrove

1 year

Serious question - What happens when we make a 1T token synthetic dataset like the snippet below and train a 100B param model on it? I really really want to know.

Tweet media one

7

1

32

@ocolegro

Owen Colegrove

7 months

@deliprao Yeah bad vibes for sure - also shocked at how many people think pplxty has their own search engine - I kept hearing "perplexity has their own index". These people believe a team of a few guys replicated decades of R&D and engineering in a year while being product focused??

3

0

64

@ocolegro

Owen Colegrove

3 years

6.) Moreover, bear market or not, the number of users in the space is only going to continue growing. The #FederalReserve isn’t solving structural issues in their monetary policy overnight. #Crypto is the future.

1

5

66

@ocolegro

Owen Colegrove

3 years

8.) As migration accelerates the technology of Avalanche, Polygon, and other EVM forks will finally be tested. Ultimately, they won’t live up to the hype.

1

5

65

@ocolegro

Owen Colegrove

7 months

RAG is not that hard to deploy, but we should aim to make it trivial. Next, we should aim to make it easier to iterate and scale. The best LLM app builders I have spoken w/ are still doing the basics. Everyone *wants* to do the sexy new thing, but in reality no one has time.

3

3

56

@ocolegro

Owen Colegrove

3 years

2.) Known supply coupled with unknown demand is a unique feature of crypto. For instance, if gold goes up in price then gold miners can use increased profits to mine more gold.

1

5

62

@ocolegro

Owen Colegrove

8 months

Everyone was right, @cursor_ai is a godsend, I feel stupid for taking so long to migrate.

10

3

62

@ocolegro

Owen Colegrove

3 years

9.). Fees will surge as thousands of users suck up all the compute with extremely expensive EVM smart contract calculations.

1

3

60

@ocolegro

Owen Colegrove

1 year

The AI Reproducibility Crisis I've been struggling of late to replicate recent findings in research that has built on top of GPT-3.5/GPT-4. I've thought about this a lot, and I believe there is growing yet largely unnoticed issue is taking root in recent AI research. I've

3

8

46

@ocolegro

Owen Colegrove

2 years

The core developers behind Algofi have accomplished a lot over the last year and it's exciting to see governance LIVE! Join the ongoing discussions at and take part in governing the largest DeFi protocol on Algorand.

4

8

58

@ocolegro

Owen Colegrove

1 year

One clear next step in the evolution of synthetic data is grounding it with RAG to reduce hallucinations and errors, e.g. "grounded synthesis". I onboarded @jerryjliu0 's @llama_index to SciPhi today to do just this. First glance results are very encouraging, RAG is the way.

Tweet media one

Tweet media two

1

3

30

@ocolegro

Owen Colegrove

3 years

There is a lot going on between this and other incentives going on in the Algorand ecosystem.

2

6

56

@ocolegro

Owen Colegrove

9 months

@abacaj You know what's even crazier? Imagine a pre-internet world. You'd either (a) have to call someone or (b) go to a literal library or book store, get a physical book, and try to look up the right answer.

6

1

57

@ocolegro

Owen Colegrove

1 year

This dataset from @laion_ai looks incredible - it won't be long before open source models are being trained on synthetic pictures as well as text.

Tweet media one

5

6

58

@ocolegro

Owen Colegrove

1 year

@yacineMTB this happens to me a lot too. I then tell them - "GPT-4 is much smarter" and they proceed to shrug and go on about their life.

2

1

54

@ocolegro

Owen Colegrove

7 months

Every good LLM prompt should probably end with `remember, brevity is the soul of wit`.

3

0

37

@ocolegro

Owen Colegrove

7 months

Cooked up a fun guide for anyone looking to run a RAG server fully locally or on-prem. Hacked it up in-flight omw back to NYC. Takes < 5m to get the server up and running with R2R, I was surprised with how quick, easy and good it is on my mac with ollama / LLama.cpp.

Tweet media one

Tweet media two

5

3

49

@ocolegro

Owen Colegrove

3 years

2) 2-3 weeks later the StableSwap will be online. With that Algofi will offer equivalent functionality to MakerDAO + Uniswap + Curve + AAVE on Ethereum

1

8

53

@ocolegro

Owen Colegrove

8 months

We added live evaluation and embedding pipeline observability to R2R today. It's starting to resemble a real product. Also, integrated @LiteLLM so that the framework can support ~any cloud LLM. Looking to have self-serve cloud with 1-click deployment in the coming week(s).

2

3

14

@ocolegro

Owen Colegrove

3 years

3.) The fixed supply prevents this from happening in crypto, and so in turn the markets are necessarily more volatile. But why does this mean that a market sell-off excites me?

1

4

51

@ocolegro

Owen Colegrove

6 months

Hard to overstate how much SciPhi benefited the last 3 months from all the start-ups I've gotten to meet from YC + the Bay - move has been very worthwhile. Impact of social networking effects in tech are significantly undervalued, imo. Explains part of EU vs. US tech imbalance

Tweet media one

6

7

48

@ocolegro

Owen Colegrove

1 year

WizardCoder vs. Phind-V2 Prelim Pass @1 Comparison Take these #'s with a grain of salt - precision comparison is not ready as the framework and my setup is evolving. Both perform similarly across 400 LeetCode problems. What we most want to see next? 8-bit quant or Llama-Code?

Tweet media one

7

4

37

@ocolegro

Owen Colegrove

10 months

@abacaj let's see, old 7b model was ~15 gb. This seems to imply that it is 8*7=56 params, so perhaps there is some kind of compression so that the size is not 8*15gb?

1

0

51

@ocolegro

Owen Colegrove

3 years

Who has tested out Zapping on the Algofi DEX TestNet? It’s still under dev, but I think it’s a great example of how interoperability can be leveraged for a better user experience - is it a feature you like?

Tweet media one

6

9

50

@ocolegro

Owen Colegrove

7 months

don't get caught in a local maxima

3

2

46

@ocolegro

Owen Colegrove

1 year

Is there an open source prompt optimizer LLM yet? This seems like something someone should train & release.

Tweet card media

DeepMind discovers that AI large language models can optimize their own prompts

Researchers from DeepMind have proposed Optimization by PROmpting (OPRO), a method that uses AI large language models (LLM) as optimizers.

venturebeat.com

9

3

42

@ocolegro

Owen Colegrove

3 years

7.). What does this all mean? That users new and old will begin migrating to the “scaleable” L2s.

2

3

48

@ocolegro

Owen Colegrove

10 months

I'm creating a webscale database of important knowledge that one can run locally. I needed this engine to build a synthetic training dataset, but it likely has some other use cases. DM if you're interested in to beta testing over the coming week or have use cases, thx!

Tweet media one

4

7

49

@ocolegro

Owen Colegrove

3 years

The Algofi team has been heads down executing for a little over 6 months now and it’s been amazing to watch the team execute and grow. @algofiorg will build enduring value alongside @Algorand . Utilizing their tech, we can build the most efficient and scaleable DeFi ecosystem.

1

10

47

@ocolegro

Owen Colegrove

10 months

The dataset is already over 4 TB and will be fully available shortly. We will be looking to proactively extend the dataset to the most important sources for LLM agents, such as SEC filings, legal documents, and more. If you have any requests, please reach out.

5

1

48

@ocolegro

Owen Colegrove

1 year

@OfficialLoganK Wormhole singularity with code flowing into the event horizon, futuristic and awe inspiring.

1

2

47

@ocolegro

Owen Colegrove

4 months

We've been building the open source ultimate RAG backend and are launching our V2! Here are some of the key features: - Configurable LLM / db provider support and streamlined builder + factory patterns for application customization

Tweet media one

2

5

38

@ocolegro

Owen Colegrove

10 months

In addition to AgentSearch, we are releasing Sensei, an LLM agent that specializes in search. Sensei was trained on a high quality synthetic data and generates truthful, grounded responses with the help of AgentSearch - .

Tweet media one

4

8

45

@ocolegro

Owen Colegrove

3 years

@AlgorandCoffee I wanted to trick some solana maxis into reading the thread :)

2

0

44

@ocolegro

Owen Colegrove

3 years

@GitGudKhed @algofiorg I will personally do some otc trades with you or anyone else that needs to convert STBL - email me at ocolegrove @gmail .com to arrange

1

2

40

@ocolegro

Owen Colegrove

2 years

I haven’t seen much discussion around the fact that the $UST implosion leaves web3 without a very decentralized stablecoin (DAI, FRAX, etc are heavily collateralized by USDC). This seems bad, what’s the likely path forward?

24

4

40

@ocolegro

Owen Colegrove

7 months

Still loving Claude Opus - it just wrote this launch post for SciPhi w/ 0 iterations. All I did was paste in ~10 recent launches from Launch YC and the first page of docs for SciPhi's open source framework and cloud offering. It might have done a better job than i could have.

Tweet media one

3

0

29

@ocolegro

Owen Colegrove

3 years

Excited for the opportunity to bring DeFi to #Algorand

@Algorand

Algorand Technologies

3 years

Today, @algofiorg announced the upcoming launch of their fast, low-cost, #decentralized lending market on the Algorand blockchain! Learn more about their #DeFi solution and why they selected Algorand over other networks 👉

Tweet media one

20

77

338

0

5

37

@ocolegro

Owen Colegrove

1 year

This is the output from an uber smol 100M param model I trained on 400M very high quality instruction tokens. Can't wait to scale it up a bit more...

Tweet media one

9

5

39

@ocolegro

Owen Colegrove

3 years

John and I will be doing an AMA at Algofest next week, tune in if you can @Algorand !

Tweet card media

From the AlgorandOfficial community on Reddit: AlgoFest 2021 Official Schedule

Explore this post and more from the AlgorandOfficial community

3

9

37

@ocolegro

Owen Colegrove

3 years

Buidl long and prosper

1

8

36

@ocolegro

Owen Colegrove

1 year

Met someone this last week who was developing their own LLM content with the intent of generating millions of pages of content to soak up Google SEO traffic. This seems like a smart way to make money tbh (for now). What's the ETA on when traditional search implodes?

16

2

38

@ocolegro

Owen Colegrove

3 years

Excellent explanation of the recent Wormhole exploit

@kelvinfichter

smartcontracts.eth (✨🔴_🔴✨)

3 years

Alright. I figured out the Solana x Wormhole Bridge hack. ~300 million dollars worth of ETH drained out of the Wormhole Bridge on Ethereum. Here's how it happened.

462

3K

13K

1

8

36

@ocolegro

Owen Colegrove

10 months

Added robust support with @bing as an optional provider for AgentSearch. Results are starting to look pretty amazing in research mode, please try it out and let me know your findings! P.S - Research mode may be a bit slow if significant traffic lands on the site tonight.

2

1

23

@ocolegro

Owen Colegrove

8 months

SciPhi will be launching 1-click cloud deployment of RAG pipelines using R2R within the next week. Think Vercel+Next.js experience for RAG. Any feature requests?

Tweet media one

Tweet media two

2

3

27