Forking Into Google’s AI Layers  🔗 **Source:** [Forking Into Google’s AI Layers](https://share.snipd.com/episode/593f70c3-ee39-4b31-b095-5d3bfa65d517) ## Highlights - **10x Science Triages AI Drug Candidates** - 10x Science is building a SaaS layer over mass spectrometry to triage thousands of AI-generated drug candidates. - They combine deterministic chemistry and AI agents to create traceable, regulator-friendly molecule analysis. Transcript: Jaeden Schafer What they're doing that's so fascinating to me is that basically there's this problem where models like DeepMind's Protein Predictor, they're spitting out thousands of drug candidates. So there's just thousands and thousands of these drug candidates. And there's a huge bottleneck in pharma where it's not just about, you know, getting all of these candidates, but it's actually triaging them. It's actually figuring out like which of all the candidates is worth pursuing to try to make medicine or therapeutics. And basically, the standard triage tool is just mass spectrometry. So it's very slow. It's very hard to interpret. It's, you know, domain experts only. 10x science is basically just building a SAS layer on top of that. And they have deterministic chemistry plus AI agents to try and make the analysis traceable and explainable, which basically matters because regulators don't accept a black box Answer on what the molecule does, right? So just because an AI model is like, oh my gosh, we discovered this super cool thing. Like this isn't going to fly. Regulators don't want that. There's no, we can't have the black box on the molecules. So we have to be able to fully understand it, fully test it. And knowing which of these, which of these chemicals, which of these drugs to test is a big problem. So everyone in the AI biotech conversation is basically talking about the generative side, and almost nobody is building the kind of picks and shovels layer that's underneath of it. And so this is why I think 10X Science is an interesting company. ([Time 0:01:01](https://share.snipd.com/snip/c0093a7c-4f9a-4e0b-9a3b-d1ecf9dd5147)) - **Anthropic Mythos Access Via Contractor Credentials** - Bloomberg reported an unauthorized group accessed Anthropic's Mythos via contractor credentials and guessed an endpoint URL pattern. - Anthropic says investigation found no evidence of impact to its own systems, only the vendor environment. Transcript: Jaeden Schafer For Anthropic today. There is a report from Bloomberg yesterday that an unauthorized group got access to Mythos, which is Anthropic's new exclusive enterprise cybersecurity AI tool that's going to take Over the world, right? They gave it to a handful of enterprises because it's so dangerous and good at getting, you know, finding security vulnerabilities. I think this is probably not good news because they, you know, they've been a lot of hype about how dangerous it was and they only give it to special people. And if it really is getting leaked, that's not great for them. I think what basically happened, if you read the reporting on Bloomberg, there was a group that operates out of a private Discord channel focused on unreleased AI models, and they got Credentials from a third-party contractor working with Anthropic. They then guessed the model's endpoint URL based on the pattern that Anthropic uses for other models, and that's it. They then gave Bloomberg screenshots and a live demo to prove access. ([Time 0:03:44](https://share.snipd.com/snip/ba8bcc7f-5699-4cad-879f-64637ea23050)) - **Audit Vendor Access To Prevent Model Leaks** - Audit vendor access and credentials for AI tools to prevent leaks via contractors. - Jaeden points to the Anthropic Mythos incident where contractor credentials and predictable endpoints enabled unauthorized access. Transcript: Jaeden Schafer It's just a contractor credential plus kind of a predictable URL pattern. And my advice that I would give to people is that if you are a company that is running AI tools, make sure that you are auditing who is on your vendor side and who has access to all those tools Right now. ([Time 0:04:51](https://share.snipd.com/snip/48a9f181-501d-42bf-a278-f876a7e127df)) - **Use System Integrators To Drive Enterprise AI Adoption** - Use system integrators like Infosys to operationalize ChatGPT for enterprise workflows. - The OpenAI-Infosys deal embeds Codex into Infosys's Topaz platform across 60+ countries for real enterprise adoption. Transcript: Jaeden Schafer I think OpenAI is trying to make a big push for enterprise right now. I mean, obviously this has been a goal for a long time, but this big push is coming right now as it feels like Anthropic is really running away with the market in enterprise. I've worked with a lot of enterprises that bought the chat GPT enterprise licenses and they kind of sat on them for like a year because they didn't wire them into anything real. This is basically the gap that I think Infosys is trying to fill. And strategically, I think this is kind of mirroring what Microsoft already does by bundling Azure OpenAI with Copilot, except Infosys plays in a lot of deals where Microsoft doesn't Always win. It's not in the deals. It doesn't have the distribution. So it's kind of a new channel. ([Time 0:06:11](https://share.snipd.com/snip/837fab57-d249-4a43-931c-fa2da087f848)) - **Google's Three Layer AI Strategy** - Google is executing a three-layer AI strategy across silicon, compute hosting, and agent interfaces. - New TPU 8T/8I chips, multi-billion deal with Thinking Machine Labs, and Chrome auto-browse tie the stack together strategically. Transcript: Jaeden Schafer Their Google Cloud Next conference kicked off today and Google did not come empty handed. They had three major announcements, TPUs, Chrome is an AI co-worker and a Thinking Machine Labs compute deal. And I think taken together, basically, they're going to make Google the most structurally positioned player in the AI stack right now. I'm going to walk through all of these different deals. But yeah, I've been pretty impressed by their announcements. And of course, we could see other players, but I would definitely say never count Google out in this industry. So first, the chips deal, Google announced two new TPUs, they have the TPU, eight T for training and the TPU eight I for inference, the split alone is I think what's really interesting To me, inference is the dominant cost of running AI and production right now and having these kind of dedicated inference silicon matters economically. It makes a big impact to the bottom line. So Google's claims is that it is three times faster at training and 80% better at performance per dollar against the NVIDIA alternatives and the ability to scale more than a million TPUs in a single cluster. So they're hitting a bunch of different things here, but NVIDIA has traditionally, I don't know if this necessarily even nvidia's fault although i'm sure some people will disagree With me here but nvidia traditionally has had a problem with um the cluster sizing and i remember when open ai's um or not opening i when xai did their first like giga terminal i don't know Whatever they're they have a robot name for it so now i can't remember but they put together 100 GP NVIDIA, 200,000 NVIDIA GPUs. It was like this big deal. They kind of figured this out. And other people started building these bigger and bigger clusters. Well, and it was kind of like they had to figure out a lot of the infrastructure and build it and think about it. Well, it looks like Google is building straight for that. And the ability to have more than a million TPUs in a single cluster is insane. And I mean, insanely scalable. So I think those are Google's own numbers. We don't have any independent benchmarks yet. Google is also not killing the NVIDIA partnership. They're reselling Vera Rubin in Google Cloud later this year. So I think basically Google wants both chips and it wants to let customers choose, which is kind of the opposite of what AWS is all in one training bet is right now, I think Google will actually Come out ahead on that. The other thing that I think is interesting that Google announced is that they're turning Chrome into an AI coworker. So they have a new feature, which is called auto-browse. Gemini powers it, and it's basically running inside of your Chrome for workspace, and it reads context across your open tabs and automates actual workplace tasks. So it can enter your CRM data, it can compare vendor quotes, it can summarize candidate portfolios, it can write up competitor research. If I'm being honest, it sounds like an interesting tool. It sounds useful. And I've actually tried some similar things. Microsoft had something like this in Microsoft Edge a while back where like all of your tabs that are open, it can just get data and read and interact with them, which is super useful. Sometimes though, you need things that are not, you know, in your tabs. And I think this is where Claude's co-work is still going to keep crushing because they're an app. You give an access to your desktop, it can get files, it can download things on your local device. So I know that like Google isn't trying to directly compete with that. They're all, they're kind of saying, you know, they're just like this human in loop, and they're not kind of letting the agents execute without your approval. And so anyways, I think like they're kind of pitching it as a as a plus, in my personal opinion, it's behind the ball, it's kind of not as good as something like Claude Cowork, which actually, Sometimes I don't want to have to get 100 pop ups asking for approval, I just wanted to get the job done, especially when I've seen it do it like three or four times successfully. I don't want to approve it. I want it to automate it. And so maybe they're playing it kind of a different target audience there, but I still would probably use Cloud Cowork over that. You can also, something that is very useful though, and kind of like Cloud Cowork, you can save frequent workflows as skills and you can fire them with a forward slash. This is ruling out in the US, workspace users are getting this first. And there's also a commitment that enterprise prompts are not going to be used to train Google's models, which I think is the right call. A lot of enterprises would be very concerned about that, especially if they wanted to put confidential data in there. But again, and not to like, I don't know, say all of the downsides to this. Again, though, you have to go and fire off a skill manually, where something like Cloud Cowork, you can just set these things up to automatically run. Okay, the next thing that I thought was interesting is that Google Cloud has just signed a multi-billion dollar deal with Miriam Moratti's Thinking Machine Labs. This is a single-digit billions. It gives Thinking Machine Labs access to NVIDIA's GB300 system on Google Cloud, plus training and deployment services for their first product, which is called Tinker, which is a tool For building custom frontier models. Thinking Machines is raising at a $12 billion valuation, and they run a heavy reinforcement learning workload, which is basically very insanely compute intensive. Mile Ott, one of the founding researchers, has basically talked a ton about how much Google got them up and running really quickly. So I think this is something that's going to be helpful for their company. The thing that I think ties all of these together is that Google is running a three layer strategy, the bottom layer, silicone, TPU, plus these resold NVIDIAs. And then they don't really care which chip wins. I mean, obviously, they kind of want their own to win, but they're going to make money either way, if you're going to be buying video from them, the middle layer is being the compute host For frontier labs. Anthropic already runs on TPUs. And now thinking machine labs is going to be running on Google cloud. The top layer is the agent layer in the browser and the workspace. Google is, you know, the only player that's hitting all three credibly right now. Anthropic is obviously crushing it with end users and, and, you know, kind of in that top layer, but they're not making their own chips, right? And so Google has a big advantage when they own the entire stack. Microsoft is really heavy on OpenAI at layer two, but really weak on silicon. Amazon is betting the farm on Anthropic. NVIDIA is the chip layer and doesn't touch the application layer. Google is basically the only company really doing the full stack. ([Time 0:07:56](https://share.snipd.com/snip/b1612299-d0ce-419a-af8d-380e0258e264)) - **TPU Claims Target Inference Cost** - Google claims TPU 8T trains 3x faster and TPU 8I gives 80% better performance per dollar for inference versus NVIDIA alternatives. - They also claim scaling to over a million TPUs in a single cluster, emphasizing inference-focused silicon. Transcript: Jaeden Schafer It makes a big impact to the bottom line. So Google's claims is that it is three times faster at training and 80% better at performance per dollar against the NVIDIA alternatives and the ability to scale more than a million TPUs in a single cluster. So they're hitting a bunch of different things here, but NVIDIA has traditionally, I don't know if this necessarily even nvidia's fault although i'm sure some people will disagree With me here but nvidia traditionally has had a problem with um the cluster sizing and i remember when open ai's um or not opening i when xai did their first like giga terminal i don't know Whatever they're they have a robot name for it so now i can't remember but they put together 100 GP NVIDIA, 200,000 NVIDIA GPUs. It was like this big deal. They kind of figured this out. And other people started building these bigger and bigger clusters. Well, and it was kind of like they had to figure out a lot of the infrastructure and build it and think about it. Well, it looks like Google is building straight for that. And the ability to have more than a million TPUs in a single cluster is insane. And I mean, insanely scalable. So I think those are Google's own numbers. We don't have any independent benchmarks yet. ([Time 0:08:43](https://share.snipd.com/snip/897aab98-8a75-4299-a428-09ed2c0bcdb5)) - **Chrome Auto Browse Adds An Agent Layer** - Chrome Auto-Browse powered by Gemini reads open tabs and automates workplace tasks while keeping human approval in the loop. - Jaeden compares it to Claude Cowork which can access desktop files and run automations more autonomously. Transcript: Jaeden Schafer So they have a new feature, which is called auto-browse. Gemini powers it, and it's basically running inside of your Chrome for workspace, and it reads context across your open tabs and automates actual workplace tasks. So it can enter your CRM data, it can compare vendor quotes, it can summarize candidate portfolios, it can write up competitor research. If I'm being honest, it sounds like an interesting tool. It sounds useful. And I've actually tried some similar things. Microsoft had something like this in Microsoft Edge a while back where like all of your tabs that are open, it can just get data and read and interact with them, which is super useful. Sometimes though, you need things that are not, you know, in your tabs. And I think this is where Claude's co-work is still going to keep crushing because they're an app. You give an access to your desktop, it can get files, it can download things on your local device. So I know that like Google isn't trying to directly compete with that. They're all, they're kind of saying, you know, they're just like this human in loop, and they're not kind of letting the agents execute without your approval. And so anyways, I think like they're kind of pitching it as a as a plus, in my personal opinion, it's behind the ball, it's kind of not as good as something like Claude Cowork, which actually, Sometimes I don't want to have to get 100 pop ups asking for approval, I just wanted to get the job done, especially when I've seen it do it like three or four times successfully. I don't want to approve it. I want it to automate it. And so maybe they're playing it kind of a different target audience there, but I still would probably use Cloud Cowork over that. You can also, something that is very useful though, and kind of like Cloud Cowork, you can save frequent workflows as skills and you can fire them with a forward slash. This is ruling out in the US, workspace users are getting this first. And there's also a commitment that enterprise prompts are not going to be used to train Google's models, which I think is the right call. A lot of enterprises would be very concerned about that, especially if they wanted to put confidential data in there. But again, and not to like, I don't know, say all of the downsides to this. Again, though, you have to go and fire off a skill manually, where something like Cloud Cowork, you can just set these things up to automatically run. ([Time 0:10:10](https://share.snipd.com/snip/495bb5b8-0f99-4900-a243-0c645e12cb98)) - **Thinking Machine Chooses Google Cloud For Frontier Training** - Google signed a multi-billion dollar deal with Thinking Machine Labs to provide NVIDIA GB300 systems on Google Cloud for training Tinker. - Thinking Machine runs heavy RL workloads and is raising at a reported $12B valuation. Transcript: Jaeden Schafer It gives Thinking Machine Labs access to NVIDIA's GB300 system on Google Cloud, plus training and deployment services for their first product, which is called Tinker, which is a tool For building custom frontier models. Thinking Machines is raising at a $12 billion valuation, and they run a heavy reinforcement learning workload, which is basically very insanely compute intensive. Mile Ott, one of the founding researchers, has basically talked a ton about how much Google got them up and running really quickly. So I think this is something that's going to be helpful for their company. ([Time 0:12:16](https://share.snipd.com/snip/45bb45ff-ca98-467f-afbe-7f966be87123)) - **Agent Integration Could Trigger Antitrust Scrutiny** - Integrating Chrome agent features with Workspace data may invite antitrust scrutiny amid Google's ongoing DOJ search remedies case. - Jaeden expects regulatory attention from the DOJ and EU shortly given the tight integration. Transcript: Jaeden Schafer The search remedies case that they've been in for a while now is still live. And turning Chrome into an agent layer that pulls from workspace data across your browser is basically the exact kind of move that makes regulators pay attention. ([Time 0:14:58](https://share.snipd.com/snip/42cfc47b-5b08-4ec8-9a31-b83cf4624e6e))