Ethan Lazuk

SEO & marketing professional.

🐹 Hamsterdam Part 67: Weekly SEO & AI News Recap (7/15 to 7/21 22, 2024)

By Ethan Lazuk

Last updated:

A weekly look-back at SEO & AI news, tips, and other content shared on social media & beyond.

Hamsterdam Part 67 weekly SEO news recap from July 15th to 21st with No Tofu quote.
Source: The Keyword

Opening notes, thoughts, and musings:

  • Welcome to another new week of Hamsterdam! 🐹
  • I appreciate you being here, and look forward to sharing the week’s news! 🙌
  • Firstly, HUGE thank you to Nicky Wake for inviting me to be a judge in the US Search Awards, along with anyone else who may have played a role in that. 🙏 (Super cool to get a LinkedIn message not pitching guest posts. 😂)
  • I rented Dumb and Dumber on YouTube last night, so I’ll try and work in some screen grabs. 🐶 🚐
  • I’m going to also try again to be briefer this week. Challenge re-accepted. 🤝
  • Fancy a longer read? Try my other blog content from this week:
  • 🇰🇭 Addressing the Gap: How the Khmer Semantic Search Engine (KSE) Works & What That Can Teach Us as SEOs 🙌
  • Bonus: Check out my TikTok finds of the week (a little bit of everything) 💎 👈 Updated!
  • Note: Not sure if this’ll load on mobile. (It’s packed.) Borrow a friend’s laptop, JIC. (JK. 😆)

We also deliver! — Subscribe to the Hamsterdam newsletter. Hot and ready in 30 minutes or less 🍕, or sent out every Sunday. (One of those two.)

If you’re in a rush, hop down to the news portion. 📰

I’m in a hurry. Take me to the news! 👇

Or continue reading for two vocabulary lessons, this week in SEO history, plus an introduction. (Don’t forget to click the dropdowns, if you want 😎.)

The Big Lebowski is this your homework Larry screenshot.

And we’re off …

🧗 Marketing word of the week: “Journey mapping”

Journey mapping is the process of visualizing your audiences’ experience with your brand from start to finish, and beyond — literally, it’s everything from their initial awareness of you to them becoming a loyal customer or viewer (and evangelist/influencer on your behalf). 🤗 🗻 🚲 🌈 ☔️ 🛣️ 🏎️👷 🚧 🚙 🛣️ 🚗 👀

Customer Journey Map from Woopra showing awareness through advocacy.
Source: Woopra.

👉 I’ve seen people reexamining the ideas around linear customer journeys or marketing funnels, partly in light of tech and generational changes. This Brand Sauce video came to mind:

@brandsauce.co

#stitch with @Ali Rose – joy & ease rethinking the marketing funnel or loop – extra spicy version (not directed at Ali but at the agency she references) #marketing #brand #brandstrategy #marketingstrategy #marketingtips

♬ original sound – joe builds brands | brandsauce

Good to roll through the comments on that video, as well. Lots of different perspectives:

TikTok comment from marketing student about using the loop.
TikTok comment about the funnel not being dead.

Of course, we’re also familiar with Think with Google’s messy middle. 🛒 🌪️

Here are a couple of screenshots from the full PDF (which you can download in that link above):

Think with Google scribble.
Just get the bare essentials Dumb and Dumber scene.
Dumb and Dumber pin wheel scene.
Messy Middle model from Think with Google paper.

Might have messily snuck in those middle two … 😆

And here’s an excerpt from a 2023 Think with Google article that elaborates on the messy middle in the context of behavioral science using Australian shopping data 🇦🇺:

“No two consumer journeys are exactly alike. A lot of factors come into play between the first trigger to purchase, right through to the purchase itself, and we know the path people take is rarely direct.

We call this in-between stage ‘The Messy Middle’. …

Our findings confirmed that marketers need to ensure they’re showing up with the right message, in the right place, at the right time — and that they can do this by leveraging key principles of behavioural science to provide consumers with the information and reassurance they need to make a decision.”

– The Messy Middle: How behavioural science can help you avoid a pricing race to the bottom (with my bolding).

🏈 There’s also a brand-focused sports marketer who I follow on TikTok named Jordan Rogers who used to work at Nike, and he once gave the tip to study the social sciences — psychology, sociology, anthropology (of course) — to understand human nature. 👈

🚨 ⬇️ 👈 Click here to read about how to create a journey map with notes specific to SEO (if you want 😎).

Here are steps Gemini gave me (with me predicting the next tokens for explanations 😆):

🤝 1. Define your target persona.

It’s common in SEO to get audience insights from keyword data, and that’s a great start for understanding topical interests or search intents. Additionally, understanding more of the audience on a personal level, like by looking at alternative sources — and using data to avoid stereotypes — can add depth to strategies, influencing how search-facing webpages or social content gets structured (UX), worded, or visualized. 🎨

🥾 2. Identify key stages of the journey.

This gets back to the first diagram we saw earlier, where the journey starts with awareness. The steps Gemini suggested are awareness, consideration, decision, and retention or advocacy. It’s pretty standard. In my experience, marketers use different lingo depending on which resources they follow, but the core idea, as I’d explain it, is that having a little something for everyone in your target audience (and potentially beyond), through all points in time and pertinent channels (organic search, paid ads, socials, email, etc.), is never a bad idea. It is, however, a bit of work and a long-term strategy. 😂

📺 3. Map touchpoints and channels.

This involves figuring out how users find your content and what they do next. On the website side, I like to put internal links wherever I think it makes sense for the next step. (👈 Captain obvious. 🧑‍✈️) I never count them or worry about multiple links to the same page. It’s literally about if it can direct someone to the next step in that micro-context. Looking broader, check out the Exploration reports in GA4 or use a tool like Clarity to see how users engage with your content, and which channels are interplaying. So often that click we see in GSC wasn’t made in isolation, nor was it the first or last interaction someone had with a brand. 🗺️

Ok, wow, Gemini listed a bunch of more steps, so I’m going to just put the rest in bullet points 😅:

  • 4. Gather data and insights – use GSC, GA4, and other analytics tools to gather and analyze relevant performance data indicative of user behavior.
  • 5. Visualize the journey map – create a visual representation of the customer journey, complete with touchpoints, channels, emotions, pain points, and opportunities for different channels (including organic search 🙌).
  • 6. Identify pain points – it never ends with ranking. What happens next counts for a lot, too. 🤗 Look for drop-off points that indicate user struggles, both micro and macro.
  • 7. Develop strategies for each stage – take each stage of the journey map into account when developing an SEO or cross-channel strategy. Different stages may require different formats of content.
  • 8. Implement your tactics – open that laptop (unless it has a blue screen 🙅, JK 😆) and put that strategy to use. 💪
  • 9. Monitor results, iterate, and refine – follow the data and seek feedback (from real users, if possible) to adjust what’s not working or take inspiration from what is – as Rick Ross once said, “Hustle, hustlin’ hustlin’.”

📚 Further reading: “Why you need Customer Journey Mapping to Boost Engagement and Conversions” by Roberta Cinus.

More Hamsterdam Marketing

😽 AI word of the week: “JAX”

“There aren’t many widely used machine learning or AI terms that start with J.” — Gemini

Tell me about it. 😂

But we’re resourceful here in Hamsterdam 🐹, so we’re going with JAX!

Duuuval chant for Jacksonville Jaguars American Football Team.
Source.

Those Floridians ☀️, I tell you what …

But no, this is JAX the machine learning framework developed by Google for high-performance numerical computing and ML research. Here’s an example:

JAX code returned in Gemini.

JAX provides a similar interface to and can extend the capabilities of NumPy, which stands for Numerical Python and is a fundamental scientific computing language — you might recall it from Jo Bergum’s talk on RAG basics.

What makes NumPy so important for scientific computing?

Well, basically, it offers a wide range of functions for doing math. 🧮 🙌

This can include trig, linear algebra (credit: al-jabr 😉), and Fourier transforms.

That last one, simply put, refers to converting a signal from the time domain (how it changes over time) to the frequency domain (what frequencies it’s made of). Yeah, it kind of confused me, too 😂, but I believe it “can reveal hidden patterns and insights, like identifying specific notes in a song,” per Gemini. 🎵 Sounds cool, anyway. 🤷

What’s important to takeaway is that many researchers and data scientists are already familiar with NumPy, so JAX allows them to leverage their existing knowledge.

What are the advantages of JAX, though?

Well, hardware accleration is one. NumPy is limited to CPUs, while JAX can leverage GPUs (like from Nvidia) as well as TPUs (Google’s Tensor Processing Units).

Large-scale ML tasks can require a lot of computation (and energy) and often parallelism (multiple calculations happening at once), making JAX handy for large-scale numerical calculations.

Another benefit is autograd. ML models often learn from training data using backpropagation and gradient descent. JAX’s built-in automatic differentiation makes the computation of gradients for cost functions easier (and faster). ⛷️

Other benefits include JIT (just-in-time) compilation, or compiling Python functions into optimized code on the fly. JAX also uses XLA (accelerated linear algebra), which means it’s efficient.

The one that stood out to me (because I actually recognized some of the vocabulary 😆) was function transformations for vectorization (vmap) and parallelization (pmap), which enable scaled computations and distributed computing. (Basically, machine learning can require a lot of resources, like multiple computers.)

No dropdown this week, but plenty of good “K” words for next week, so expect a full report then. 🫡

More Hamsterdam Research

This week in SEO history: Early web image (1992)

I often find SEO-related history to reference on Web Design Museum. 💾 👾 🛜

This week, I came across “Les Horribles Cernettes,” one of the first images on the web. 🙌

Les Horribles Cernettes, one of the first images on the web, via Web Design Museum.
Source.

“Silvano de Gennaro, an Italian computer scientist who worked at CERN research labs, was asked by Tim Berners-Lee to scan and upload a photo of a parody pop-group called Les Horribles Cernettes (The Horrible CERN Girls) onto the info.cern.ch website. This photo became one of the first images to be published on the World Wide Web.“

– Les Horribles Cernettes, one of the first image on the Web (Web Design Museum)
Google desktop AI Overview of Les Horribles Cernettes.

🚨 ⬇️ 👈 Click to view the sources in the above image (if you want 😎).
More Hamsterdam History

🎙️ Introduction to week 67: Inspired stories

Hearing about other’s stories often inspires me. I hope mine below does the same for someone.

I’ve told this before in other places, like on my about page.

But this is the fuller version …

… of a part …

… of how I got into SEO.

I was a grant writer at Goodwill Industries of Southwest Florida in Fort Myers — great golf town. 🏌️

It was pretty cool, except when it wasn’t. 😂

👉 But the realest reward was seeing my colleagues — most of whom had left behind higher-paying roles — delivering services to people with disabilities or of lower income, sometimes receiving gratitude, but other times not.

As a grant writer, I mostly just described what those people actually did, but I sometimes got to support the programs, too.

The more money you’re going after, the more red tape for your grants. (Some of it is relationships, too.)

In fact, someone at a big-giving organization once told me they get so many grant applications that they’d just toss them out for any small infraction. 🗑️

Literally. It could be font size, attaching a Word doc instead of a PDF, etc.

One person even took the time to add a handwritten note to their rejection letter telling me they don’t give money to Goodwill Industries because they don’t like it. Not sure if they knew every Goodwill is independent. 🤷

Anyway, if you’re writing a grant, don’t let that be you. It sucks. 😆 (Stay tuned.)

Here’s another thing to consider …

The people who ran the programs never wrote the grants, or even looked at them. They were too busy.

Their only involvement came when I needed to get some info or a random stat.

That’s not true of all organizations, of course. Smaller nonprofits’ team members may wear more hats. 🧢

This one time, we were trying to get a grant from, I think, Home Depot to build a wheelchair ramp for a Vietnam veteran.

I put my heart and soul into that grant. I also filled out the wrong one. 😳

There was a panicked dialogue behind the scenes.

Dumb and Dumber screen grab of our pets' heads are falling off scene.

Fortunately, I was able to write the correct grant application and submit it the night of the deadline.

But wow, it made me think about how we’re all part of this big continuum, and someone’s good or bad day, right or wrong move, can translate itself 50 miles (80.4672 kilometers 🤘) down the road.

Anyway, a recruiter called me one day out of the blue and asked if I wanted to join a digital marketing agency in downtown Fort Myers as a content specialist.

Low stakes, so I had fun with my writing sample, they dug it 🤝, and the rest, as they say, is history. 😂

Buckle up for a full week’s recap, and enjoy the vibes (which I remember watching live on MTV as a 14-year old 🔥):

Avril Lavigne Fuel Metallica cover YouTube comment.
Comment from Fuel video. 😂

Thank you for supporting Hamsterdam and the cause of SEO & AI learning. 🙏

Missed last week? Don’t worry, I got you! Read Part 66 to catch up.

🌟 Other great sources of weekly SEO news:

Time for our weekly review of SEO social posts, articles, & more …

⚡️ Quick summary:

  • Google core update dropping soon (or maybe has, I’m taking a short break from socials 😂); Danny Sullivan bravely goes “Where no Search Liaison has gone before” 🛸 (not sure if that last part is true)
  • Google Search Console shares Arabic resources
  • AIOs going more global in the UK
  • 🔥 Pick of the week: AI Has Changed How Search Works by Marie Haynes, SEJ (extract is from SEO in the Gemini Era by Marie Haynes ©2024) – MUST READ, and I especially like this part, this part and this part
  • 🐿️ Sneaky pick of the week: Marketing Expert Breaks Down 8 Mad Men Pitches by Hiten Shaw – This is a really cool idea/video; love me some Mad Men:
Peggy Olson Mad Men Last Episode Walk.

⏩ Jump to a section:

Or keep scrolling to see it all. ⏬

Now, let’s step inside the white flags of Hamsterdam …

The Wire Hamsterdam screenshot for setting up inside the white flags.

📰 SEO news, Google updates, SERP tests & notable posts

Notable updates or news related to Google Search or related SEO topics (aka, the Barry section).

🫡

🍟 SEO tips & tidbits

Actionable tips, cool tidbits, and other snackable findings and observations that can be teaching moments.

Marianne Sweeny LinkedIn post about ontologies.
Read Marianne’s post on LI.
LOVE this. 👆 Nothing worse than intellectual theft. Nothing better than a peer-to-peer shoutout. 🤗
Miriam Ellis LinkedIn post about her new journey and services.
Read Miriam’s post on LI.

🦕 SEO (and AI) fundamentals & resources

Essential information, concepts, or resources to learn about SEO or AI.

Excerpt: “The person asking the question commented that the SEO Starter Guide recommends using heading elements in ‘semantic’ order for people who use screen readers (devices that translate text into spoken words) but that otherwise it’s not important for Google. The person asking the question wanted to know if the SEO Starter Guide was out of date because an SEO tool had a different recommendation.”
Excerpt: “While acknowledging that increased engagement requires additional time from both parties, Mueller believes it’s worth the effort.”
Excerpt: “So much of what many of us do as SEOs and treat as standard practice is based on a search engine that was a list of heuristics – handwritten rules programmed by humans. So much has changed.”

📚 Articles, videos, case studies & more

Longer-form content pieces shared on social, in newsletters, and elsewhere.

Excerpt: “The stated goal for Google’s algorithm decisions is to increase user satisfaction but the problem with that approach is that website publishers are left out of that equation. Consider this: Google’s Search Quality Raters Guidelines says nothing about checking if big brands are dominating the search results. Zero. Website publishers aren’t even an afterthought for Google. Publishers are not not considered at any stage of the creation, testing and rollout of ranking algorithms.”
Excerpt: “The DMA does something very important for anybody in search: it imposes a no-preferencing rule on gatekeepers. In the case of Google specifically, this essentially means that Google’s own vertical search services – such as Google Shopping and Hotels – cannot be preferred over other vendors. Now, this is a little too late to help the companies that Google already destroyed with their search monopoly (i.e. the shopping comparison engines that were destroyed by Google Shopping). BUT it’s not too late for OTAs and aggregators in the travel space.”
Excerpt: “Both Baidu and Google have embraced machine learning to improve search capabilities. These algorithms have significantly enhanced Baidu’s ability to understand intent, analyze content and deliver personalized search experiences. Baidu leverages machine learning across various aspects of search, including ranking algorithms, content understanding, multimedia analysis, voice search processing and anti-spam measures. This shift underscores the need for high-quality, user-focused content that aligns with search intent.”
Excerpt: “Blog posts, X posts, YouTube descriptions or Pinterest posts – everything is suddenly a potential piece of content that might be used as training material for a large language model … User behavior has changed dramatically, making it impossible for brands to stay focused on just one channel and ignore the rest. AI does not pick a channel.”
Excerpt: “Typically, we can expect to see a 30% to 40% decline in traffic coming to websites, depending on the vertical, due to AI Overviews and zero-click rich results. While these results reference similar sources and domains used for rich snippets and organic results, they work on a fundamentally different underlying system.”
Excerpt: “A lot of organizations often debate whether or not they should or shouldn’t invest in G2, but the reality is that G2 is being used by B2B buyers every single day. A lot of Savvy marketers, influential tech heads, and gurus will say, ‘no one actually uses G2,’ but the numbers don’t lie. The numbers suggest that G2 is still very much a key factor in driving decisions for B2B buyers and that SaaS brands ignoring the channel might actually be missing out.”

🧑‍💻 Technical SEO

Everything from basics to advanced moves (and also tools).

Excerpt: “Interaction to Next Paint measures how responsive a page is to visitor interactions. It measures the elapsed time between a tap, a click, or a keypress and the browser next painting to the screen. INP breaks down into three sub-parts.”

✍️ Content marketing

From what is helpful content to user journeys and beyond.

Writing Description Part 1 – Lynette Eason, The Steve Laube Agency

Writing Description Part 1 by Lynette Eason, The Steve Laube Agency.
Excerpt: “Readers need to be anchored into place. They need to know where the characters are because that’s where the reader will be. So, some description establishes the time and place of the story.”

📍 Local SEO

From Google Business Profiles or reviews and more!

📊 Data analysis & reporting

Showing that what you’re doing is helping.

Excerpt: “Understanding the languages that Google is automatically translating can help site owners and SEOs determine which languages they might want to translate their content to for users. For example, you might want to have someone fluent in that local language translate your content, and then use hreflang to signal to Google the language and country combinations (for swapping in the right content by language and country in the search results).”
Daniel Foley Carter LinkedIn post about reporting tools.
Read Daniel’s post on LI.

🤖 AI, machine learning, & LLMs

News related to models, papers, and companies.

Excerpt: “OpenAI CEO Sam Altman has made clear he wants to disrupt the search paradigm via a new search product that combines web search plus large language models (LLMs). ChatGPT already offers an impressive “browse mode” to search the web that got really good following the arrival of GPT-4o.”
Excerpt: “The danger of anthropomorphism, I think, is in thinking that a system, such as a large language model, chatbot or something, is thinking that it has capabilities and that it doesn’t. It’s as simple as that. Actually, it’s also thinking perhaps that it lacks the capability that it does. So in both cases, I think we can go wrong. You can go wrong because they exhibit very human-like linguistic behavior, we can just assume that they are going to be very human-like, in general, in all of the rest of the behavior that we encounter with them.” (From show notes.)

Mistral releases Codestral Mamba for faster, longer code generation – Emilia David, VentureBeat

Mistral releases Codestral Mamba for faster, longer code generation by Emilia David, VentureBeat.
Excerpt: “Mamba seeks to improve upon the efficiency of the transformer architecture used by most leading LLMs by simplifying its attention mechanisms. Mamba-based models, unlike more common transformer-based ones, could have faster inference times and longer context. Other companies and developers including AI21 have released new AI models based on it.”

What happened to BERT & T5? On Transformer Encoders, PrefixLM and Denoising Objectives – Yi Tay

Excerpt: “A quick primer (Skip connection to next section if you feel confident): There are mainly three overarching paradigms of model architectures in the past couple of years. Encoder-only models (e.g., BERT), Encoder-Decoder models (e.g., T5) and decoder-only models (e.g., GPT series). People get confused a lot about this and people often have tons of misconceptions about these dichotomies and architectures so I’m hoping this post will help. The first thing to really understand is that encoder-decoder models are actually still autoregressive models. A decoder in an encoder-decoder model is literally and fundamentally still a causal decoder. Instead of pre-filling a decoder model, some text can be offloaded to an encoder, which is then sent to the decoder via cross-attention. Yes, T5 models are also language models!”

The framework helping devs build LLM apps – Eira May, Stack Overflow

“Ben and Eira talk with LlamaIndex CEO and cofounder Jerry Liu, along with venture capitalist Jerry Chen, about how the company is making it easier for developers to build LLM apps.”:

The framework helping devs build LLM apps by Eira May, Stack Overflow.
Excerpt: “JL: So taking a step back, I think there’s a lot of different companies building RAG systems over different sources of data, and there’s also a lot of different data types. As you mentioned, there’s all the workplace apps developer tools that you might use. Another popular data source is just buckets of unstructured files for instance, like PDFs, powerpoints, and then there’s also, of course, structured data like data warehouses. … We started off as an open source developer toolkit that emerged into a framework of just tools for different developers to use and compose their own applications. Of course there’s a lot of just general high-level out of the box search tools, whether it’s internal company search or code search or anything, that have emerged that basically directly are targeted to the end user. But our entire belief is that at the developer level, a lot of developers are going to be leading the charge of basically enabling and spreading Gen AI adoption throughout the enterprise.” (Via the transcript.)

🤔 General marketing & miscellaneous

This is for great content that isn’t necessarily SEO or marketing-specific. PPC, PR, dev, design, and social friends, check it out!

Three Keys to Mindfulness: Awareness, Attitude, and Values – Willem Kuyken Ph.D., Psychology Today

Three Keys to Mindfulness: Awareness, Attitude, and Values by Willem Kuyken Ph.D., Psychology Today.
Excerpt: “‘My mind can be my worst enemy or best friend,’ said Raheem Sterling, a young English soccer player. Sterling is known for inspiring others through his work ethic, for his family values, and for raising awareness of important social issues. He meditates because it helps him befriend his mind, which in turn helps him with his sport, mental health, and wider work.”

10 Tips to Supercharge Your Brand Like Taylor Swift – Nuala Walsh (CEO, MindEquity) for Inc.

10 Tips to Supercharge Your Brand Like Taylor Swift by Nuala Walsh (CEO, MindEquity) for Inc.
Excerpt: “Tell Authentic Stories: Like Sinéad O’Connor, Swift’s songs reflect struggles or unrequited romantic experiences, making her more human and relatable. To deepen connection, savvy businesses can focus on narratives that reflect their values and their customers’ stories. For instance, Dove has built its brand around body positivity through its ‘Real Beauty’ campaign. It feels real.”
Excerpt: “Instead, they got started on AlphaFold 3. This newer model, which the teams at Google DeepMind and Isomorphic Labs launched in May, builds on our previous models by predicting not just protein folding structures, but predicting the structure and interactions of all of life’s molecules, including DNA, RNA and ligands (small molecules that bind to proteins).”
Excerpt: “Happy World Emoji Day! You didn’t think we’d let such a momentous occasion pass without a bit of fanfare 🎉, did you 😉? And there’s a lot to celebrate. For starters, new emoji are on the way. To mark the day, we’re sharing more details on that tidbit, plus nine more fun facts about emoji — you might just learn a thing or 2️⃣.”
Excerpt: “Merchant Center Next (GMC Next) brings a revamped interface and new features to Google’s ecommerce platform, potentially streamlining operations for online retailers.”

The Power of Your Personal Strengths – Diane E Dreher Ph.D., Psychology Today

The Power of Your Personal Strengths by Diane E Dreher Ph.D., Psychology Today
Excerpt: “In our challenging world today, too many of us are languishing. Across the country and around the world, there are alarming rates of anxiety, loneliness, and depression (Murthy, 2023; World Health Organization, 2024). … Centuries ago, in the wake of the bubonic plague pandemic, people began believing in themselves and their strengths, which led to the unprecedented creative flourishing of the Renaissance.”

💎 Older stuff that’s good!

Not everything I find worth sharing is new as of this week, so these are gems I came across published in the past.

Deep Dive into Vector Databases by Hand ✍︎ – Srijanie Dey, PhD, Towards Data Science

Deep Dive into Vector Databases by Hand by Srijanie Dey, PhD, Towards Data Science.
Excerpt: “The other day I asked my favorite Large Language Model (LLM) to help me explain vectors to my almost 4-year old. In seconds, it spit out a story filled with mythical creatures and magic and vectors. And Voila! I had a sketch for a new children’s book, and it was impressive because the unicorn was called ‘LuminaVec’.”

Temperature Scaling and Beam Search Text Generation in LLMs, for the ML-Adjacent – Mike Cvet, Towards Data Science

Temperature Scaling and Beam Search Text Generation in LLMs, for the ML-Adjacent by Mike Cvet, Towards Data Science.
Excerpt: “Since we’re working with natural language, the complexity involved is far too high to exhaust the search space for every query in most contexts. The solution is to trim that search space down to a reasonable number of candidate paths through the candidate token graph; maybe just 4, 8, or 12. Beam search is the heuristic generally used to approximate that ideal A*-like outcome. This technique maintains k candidate sequences which are incrementally built up with the respective top-k most likely tokens. … Now, what does temperature have to do with all of this? As I mentioned above, this parameter doesn’t really inject randomness into the generated text sequence, but it does modify the predictability of the output sequences. Borrowing from information theory: temperature can increase or decrease the entropy associated with a token prediction.”

Great job making it to the end. You rock! 🪨

🤝 Want help with your SEO strategy?

I’m an independent SEO consultant focusing on custom audits and holistic strategies for brands. Don’t hesitate to reach out, or visit my about page for more information.

Let’s connect!

Hit me up anytime via text or call at 813-557-9745 or on social or email:

Cheers! ✌️

SEO Strategist and The Wire Aficionado

Ethan Lazuk

Editorial history:

Created by Ethan Lazuk on:

Last updated:

Need a hand with a brand audit or marketing strategy?

I’m an independent brand strategist and marketing consultant. Learn about my services or contact me for more information!

Leave a Reply

Discover more from Ethan Lazuk

Subscribe now to keep reading and get access to the full archive.

Continue reading

GDPR Cookie Consent with Real Cookie Banner