Skip to content

Commit

Permalink
Updated site
Browse files Browse the repository at this point in the history
  • Loading branch information
jonah-ramponi committed Aug 23, 2024
1 parent 208a618 commit 3f511c8
Show file tree
Hide file tree
Showing 3 changed files with 10 additions and 10 deletions.
2 changes: 1 addition & 1 deletion index.xml
Original file line number Diff line number Diff line change
Expand Up @@ -13,7 +13,7 @@
<link>https://www.jonahramponi.com/posts/meta_analysis/</link>
<pubDate>Fri, 23 Aug 2024 00:00:00 +0000</pubDate>
<guid>https://www.jonahramponi.com/posts/meta_analysis/</guid>
<description>To understand why Meta has open sourced the Llama family of models, I think it is important to understand how Meta makes money. Meta makes money from adverts. Almost their entire revenue comes from adverts [1]. So, why have Meta invested so much money into the Llama models?&#xA;Probably, to make more money from adverts.&#xA;Here are some ways in which the open source release of the Llama models might help Meta make more money from adverts.</description>
<description>To understand why Meta has open sourced the Llama family of models, I think it is important to understand how Meta makes money. Meta makes money from adverts. Almost their entire revenue comes from adverts (1). So, why have Meta invested so much money into the Llama models?&#xA;Probably, to make more money from adverts.&#xA;Here are some ways in which the open source release of the Llama models might help Meta make more money from adverts.</description>
</item>
<item>
<title>Intro to Attention</title>
Expand Down
2 changes: 1 addition & 1 deletion posts/index.xml
Original file line number Diff line number Diff line change
Expand Up @@ -13,7 +13,7 @@
<link>https://www.jonahramponi.com/posts/meta_analysis/</link>
<pubDate>Fri, 23 Aug 2024 00:00:00 +0000</pubDate>
<guid>https://www.jonahramponi.com/posts/meta_analysis/</guid>
<description>To understand why Meta has open sourced the Llama family of models, I think it is important to understand how Meta makes money. Meta makes money from adverts. Almost their entire revenue comes from adverts [1]. So, why have Meta invested so much money into the Llama models?&#xA;Probably, to make more money from adverts.&#xA;Here are some ways in which the open source release of the Llama models might help Meta make more money from adverts.</description>
<description>To understand why Meta has open sourced the Llama family of models, I think it is important to understand how Meta makes money. Meta makes money from adverts. Almost their entire revenue comes from adverts (1). So, why have Meta invested so much money into the Llama models?&#xA;Probably, to make more money from adverts.&#xA;Here are some ways in which the open source release of the Llama models might help Meta make more money from adverts.</description>
</item>
<item>
<title>Intro to Attention</title>
Expand Down
16 changes: 8 additions & 8 deletions posts/meta_analysis/index.html
Original file line number Diff line number Diff line change
Expand Up @@ -96,25 +96,25 @@ <h1 class="title">Why did Meta publish the Llama models for free?</h1>
</div>

<section class="body">
<p>To understand why Meta has open sourced the Llama family of models, I think it is important to understand how Meta makes money. Meta makes money from adverts. Almost their entire revenue comes from adverts [1]. So, why have Meta invested so much money into the Llama models?</p>
<p>To understand why Meta has open sourced the Llama family of models, I think it is important to understand how Meta makes money. Meta makes money from adverts. Almost their entire revenue comes from adverts (1). So, why have Meta invested so much money into the Llama models?</p>
<p><em>Probably, to make more money from adverts.</em></p>
<p>Here are some ways in which the open source release of the Llama models might help Meta make more money from adverts.</p>
<h3 id="to-increase-the-value-of-their-data">To increase the value of their data</h3>
<p>Open sourcing the Llama model weights has led to massive adoption of the Llama models by developers and researchers. This community-driven effort has greatly advanced understanding of generative AI models. These advancements should enable Meta to extract more value from their data.</p>
<p>A while back, a leaked memo from a Google Researcher titled <em>“We Have No Moat, And Neither Does OpenAI“</em> was released [2]. Meta promptly decided to get into the bridge building business with the open source community . They do, however, keep a deep moat filled with crocodiles around what is truly valuable to them - their data.</p>
<p>A while back, a leaked memo from a Google Researcher titled <em>“We Have No Moat, And Neither Does OpenAI“</em> was released (2). Meta promptly decided to get into the bridge building business with the open source community . They do, however, keep a deep moat filled with crocodiles around what is truly valuable to them - their data.</p>
<p>It would not surprise me if we never hear about the models trained on our Instagram and Facebook data. I think it will be these models which will reap the biggest rewards for Meta - because those models might increase advertising revenues by a percentage or two, and that is worth a lot.</p>
<p>So what exactly might this value add look like?</p>
<p>I don’t know. That’s a question for the researchers at Meta under super strict NDAs. But maybe:</p>
<ul>
<ol>
<li>
<p>Improvements to multimodal models could allow them to better extract information from user’s posts. For instance, they could better identify the types, values, styles, etc., of clothes an individual wears. They might better understand the look they go for. This could be used for targeted advertising.</p>
</li>
<li>
<p>Improvements to language models may allow for better analysis of the tone and sentiment in posts. Meta could then better tailor ads to match the user’s current emotional state, improving engagement and relevance. For instance, if emotional stress were detected Meta could recommend using Facebook less.</p>
</li>
</ul>
</ol>
<p>But why didn’t Meta just train the Llama models privately and keep them for internal use? Why expose themselves to potential legal issues or lawsuits by making them open source?</p>
<p>I believe the prevailing theory is that Meta is strategically commoditizing its complements [3]. They have a core revenue stream—advertising—and by making certain technologies widely available, they indirectly enhance their primary money-making activity.</p>
<p>I believe the prevailing theory is that Meta is strategically commoditizing its complements (3). They have a core revenue stream—advertising—and by making certain technologies widely available, they indirectly enhance their primary money-making activity.</p>
<p>I would guess that Meta intends to commoditize as much of their tech stack as possible. For instance, perhaps they might create a new frontend framework which makes it easier to build scalable social network sites. Or a better framework than tensorflow for building digital advertising ML models.</p>
<p>Meta might initial development to what they need internally, providing top quality engineering for the initial codebase. Once released to the open-source community, developers around the world contribute to refining and expanding its functionality.</p>
<p>When a company like Meta finds itself behind the competition in certain areas, open-sourcing its work can help bridge the gap. This approach ensures that Meta not only catches up thanks to continual improvements by the community. In turn, these developments feed back into Meta’s ultimate goal—better advertising revenues</p>
Expand All @@ -132,9 +132,9 @@ <h3 id="because-big-tech-is-a-battlefield">Because big tech is a battlefield</h3
<p>There were many other reasons Facebook wanted Waze; another key reason would be to deliver better location-based mobile advertising and optimise local search results. Right before they could buy Waze for 500m$, Google stepped in and spent 1bn$ to snatch it. Why? Google bought Waze so that Facebook would not have it. Google probably didn&rsquo;t really need Waze, but it partially contributed to Facebook abandoning their smartphone plans. It stunted Facebook&rsquo;s expansion. For big tech, messing up each other&rsquo;s game is a legitimate business strategy.</p>
<p>Similarly, Microsoft&rsquo;s (and OpenAI’s) dominance in Generative AI might have been enough of a threat to prompt Meta into action. By making AI tools freely available, Meta isn’t just playing nice—it’s playing smart. It’s a calculated move to dilute Microsoft’s power and level the playing field, ensuring that no one company holds too much sway in the AI landscape.</p>
<p>So, while Meta’s actions might seem altruistic on the surface, they’re likely driven by the same competitive instincts that have shaped big tech for years. Whether this leads to a better world, with more innovation and access, or just a noisier one full of whoopee cushions, remains to be seen.</p>
<p>[1] In the second quarter of 2024, 98.1% of Meta’s revenue came from adverts. <a href="ps://investor.fb.com/investor-news/press-release-details/2024/Meta-Reports-Second-Quarter-2024-Results/default.aspx">Meta Q2 Earnings</a>
[2] <a href="https://www.semianalysis.com/p/google-we-have-no-moat-and-neither">We Have No Moat, And Neither Does OpenAI</a>
[3] <a href="https://www.joelonsoftware.com/2002/06/12/strategy-letter-v/">Joel on Software discusses Commoditizing your Complement</a></p>
<p>(1) In the second quarter of 2024, 98.1% of Meta’s revenue came from adverts. <a href="ps://investor.fb.com/investor-news/press-release-details/2024/Meta-Reports-Second-Quarter-2024-Results/default.aspx">Meta Q2 Earnings</a>
(2) <a href="https://www.semianalysis.com/p/google-we-have-no-moat-and-neither">We Have No Moat, And Neither Does OpenAI</a>
(3) <a href="https://www.joelonsoftware.com/2002/06/12/strategy-letter-v/">Joel on Software discusses Commoditizing your Complement</a></p>

</section>

Expand Down

0 comments on commit 3f511c8

Please sign in to comment.