Correct blog post typos (#456)

Co-authored-by: pdamodaran <pdamodaran@users.noreply.github.com>

Author: daniellemaxwell

content/en/blog/2024-02-01-how-to-plan-and-scope-llm-projects.md

@@ -105,7 +105,7 @@ Estimate to complete each task in 1 to 2 sprint cycles, with a little cushion ti
 - **First-party data versus third-party data:** First-party data will be cheaper than third-party data and often that might be all that is needed to develop the solution.  Only consider augmenting with third-party data if the data that is present within the organization is insufficient.  How much data do you need?  The answer is not as much as you may think.  Refer to the example below.
 - **Computer hardware requirements:** If you are fine tuning an open source LLM on a small subset of your data, it is entirely possible to have a working solution using just a single beefy laptop.  In other words, it is not a requirement to load up on GPUs.
 
-*To wrap all of this into a specific example, we were involved in a project where we built a domain-specific LLM for an organization using a pre-built model from Hugging Face using just 12 MB of the company's data and through the course of **three iterations** built a solution in a single laptop that **revolutionalized** the organization's business.*
+*To wrap all of this into a specific example, we were involved in a project where we built a domain-specific LLM for an organization using a pre-built model from Hugging Face using just 12 MB of the company's data and through the course of **three iterations** built a solution in a single laptop that **revolutionized** the organization's business.*
 
 
 ## Bonus Tip: Stay Agile

content/en/blog/2024-02-06-build-your-own-llm---training.md

@@ -38,7 +38,7 @@ Note that some models only an encoder (BERT, DistilBERT, RoBERTa), and other mod
 Training an LLM from scratch is intensive due to the data and compute requirements. However, the beauty of Transfer Learning is that we can utilize features that were trained previously as a starting point to train more custom models. More specifically, fine-tuning is the process of using a model that has been exhaustively _pre-trained_ and continuing the training with a custom data set. Theoretically, we should be able to take a large pre-trained model [like distilbert-base-uncased](https://huggingface.co/distilbert/distilbert-base-uncased) and train it on the movies dataset we ingested in [part one](https://rotational.io/blog/build-your-own-llm---data-ingestion/). The goal is to train a model that:
 
 1. Better "understands" the domain of movie reviews.
-2. Has additional layers at the end to classify reviews as postive or negative.
+2. Has additional layers at the end to classify reviews as positive or negative.
 
 Note: We're using the _uncased_ version of distilbert which treats cases the same (e.g. `Ryan Gosling` == `ryan gosling`). You can also try training from the [cased](https://huggingface.co/distilbert/distilbert-base-cased) version and see how it impacts the resulting model.
 

content/en/blog/2024-11-06-hmac-verification-tokens.md

@@ -139,7 +139,7 @@ func (t Token) String() string {
 }
 ```
 
-The message is marshalled into a binary form, then a rand secret key is generated and used to create the HMAC signature along with a SHA256 hash. This signature is stored alongside the message and the ID and the secret are concatened together to send to the recipient.
+The message is marshalled into a binary form, then a rand secret key is generated and used to create the HMAC signature along with a SHA256 hash. This signature is stored alongside the message and the ID and the secret are concatenated together to send to the recipient.
 
 You should store the `SignedMessage` in your database for retrieval later. I use the `Scanner` and `Valuer` interface for this, storing a binary representation of the signed message as a BLOB. This allows me to include the `nonce` and the `signature` in the binary data without worrying about exporting those fields. Just remember, you need to index on the `ID` field, so store that separately!
 

content/en/blog/2024-11-27-product-development-in-the-ai-era.md

@@ -11,7 +11,7 @@ tags: ['AI', 'ML', 'Strategy']
 description: "Get insights from seasoned experts on how to build successful AI products"
 ---
 
-Last month, we spoke to a panel of expert product managers who have successully implemented AI solutions in their organizations.  This post recaps some of the key strategies and insights they shared.      
+Last month, we spoke to a panel of expert product managers who have successfully implemented AI solutions in their organizations.  This post recaps some of the key strategies and insights they shared.      
 
 <!--more-->
 
@@ -48,7 +48,7 @@ Evelyn favors learning by doing as opposed to getting the latest certification.
 
 ## What AI features excite them the most
 
-When asked what AI features excite them the most, Jessica mentioned drug discovery applications.  She added that there are a lot of opportunities in many areas such as DevOps where AI can be used to manage costs and to speed up deployments.  Sebastian quipped that he would love to have AI create slide decks for him.  Evelyn took it one step further and is excited about the product that she is building that will streamline board meetings and quaterly business reviews, which includes creating the slides and cleaning up the data.  Max is excited about agents, including the product that his team built that summarized a dashboard that saved 30 minutes of time by removing the need to answer rudimentary questions.
+When asked what AI features excite them the most, Jessica mentioned drug discovery applications.  She added that there are a lot of opportunities in many areas such as DevOps where AI can be used to manage costs and to speed up deployments.  Sebastian quipped that he would love to have AI create slide decks for him.  Evelyn took it one step further and is excited about the product that she is building that will streamline board meetings and quarterly business reviews, which includes creating the slides and cleaning up the data.  Max is excited about agents, including the product that his team built that summarized a dashboard that saved 30 minutes of time by removing the need to answer rudimentary questions.
 
 ## Final Thoughts
 - **Jessica:** Get in there and play.  Lead from strategy.  Work on storytelling and influence.

content/en/blog/2024-12-05-rotational-joins-eu-us-dpf.md

@@ -23,7 +23,7 @@ At Rotational Labs, data drives our business, and always will. As we develop and
 
 While our internal efforts and policies are intentional and effective, we recognize the growing importance of making a public commitment to responsible data stewardship on a global scale.  
 
-To reinforce our committment to privacy, Rotational has joined the **[EU-US Data Privacy Framework (DPF)](https://www.dataprivacyframework.gov/)**, a transcontinental initiative sponsored by the U.S. Department of Commerce and its counterparts in the EU and UK. The DPF establishes rigorous data protection policies, promotes responsible data use, and ensures independent recourse for dispute resolution.  
+To reinforce our commitment to privacy, Rotational has joined the **[EU-US Data Privacy Framework (DPF)](https://www.dataprivacyframework.gov/)**, a transcontinental initiative sponsored by the U.S. Department of Commerce and its counterparts in the EU and UK. The DPF establishes rigorous data protection policies, promotes responsible data use, and ensures independent recourse for dispute resolution.  
 
 ### Privacy in the Age of Generative AI