DEV Community: Micah Wade

How X AI Replaced Google as The Third Best AI Lab

Micah Wade — Fri, 17 Jul 2026 00:00:00 +0000

Google's failures

As someone who uses Antigravity CLI for coding, I have to say it's not the best model and has a huge amount of issues. With Gemini 3.5 Flash dropping 3 months ago and with no release date on 3.5 Pro, it seems like Google is falling behind. The current models use way too many tool calls for simple changes and don't stop after they have solved the problem. These two issues mean that Gemini models burn more tokens or credit for a task compared to other models.

If we look at benchmarks overuse of tokens seems to be an issue with Google models. In DeepSWE benchmark V1.1 Gemini models are at the bottom right which makes them the least efficient models that don't even score well.¹ The insane thing is the top model GPT 5.6 Sol on Max is cheaper and scores the best of any model.

What you need to make LLMs

To train a good large language model you need three main parts: data, compute and researchers. While that is not everything you need it's the main parts you need. Data is not all created equally. There are two types of data, first is general and world knowledge which is used while doing the main training of the model.

Once you have trained the base model you can start on reinforcement learning at this stage there are two important types of data for a good LLM at this stage. Chat histories don't have to be directly code but even chat's about a semi aquatic mammal are needed. Chat histories about non code may not be need if we are looking at just code performance right? You need to remember that all LLMs are is text prediction machines. This means that it needs to be fine tuned to work better in a chat setting.

Now direct code prompts and answers can be very good because LLMs don't know how to deal with tool calls. This data is where a model learns to make tool calls and with current frontier models call sub agents to investigate and solve small problems. This type of data is also important for teaching how it should write code like comments and style.

Now we have all this data how do we turn it into a large language model? We need computers to put all this data into the models weights but not all compute is equal. CUDA has become the default for training LLMs with most if not all models trained in CUDA. While Nvidia basically has a monopoly with CUDA on training. But for inference there is a bit more variety which I may talk about in the future.

Researchers have become a huge asset to companies with Researchers being given 1M+ packages to join companies. With Researchers becoming more expensive companies have been forced to get them Nvidia GPUs as they are easier to work with.

What Google has

Before the huge AI hype train Google had one of the best AI Researcher talent pool. This huge pool of talent was mostly gained by their acquisition of DeepMind in 2014.² After the acquisition it was merged with Google Brain to form one subsidiary. From an outsiders point it looked like Google left them alone, which for the researchers is probably what they wanted. Now inside of DeepMind, Google probably has its hands all in there trying to get the next model released with it being a high stress place to work. This can be seen with departures of prominent researchers. With this talent leaving it is most likely becoming a more stressful place to work which then makes more researchers leave and then it perpetuates.³

Now let's talk about data while Google is known as the advertising company that wants your data this data is not what you need to train an LLM. For general data I think Google has a good amount of this data due to scraping the internet. For the second type of data Google has strong areas but also weak areas. Chat history is where Google should have a good amount of data with the Gemini app having 750 million active users with training on history on by default.⁴ Now let's talk about the type of data Google has the least of Coding. Google does have Antigravity but due to the worse code quality of the current models there is not enough people using it in order to get the data they need. Now I think Google is trying to solve this because now Gemini can create simple code that can be used to make interactive things in the Gemini App which will help them gather the data they need.

Compute is where Google has an advantage which has disadvantages. No other top AI lab has their own silicon but Google does with their custom tensor processing units. While this provides a way to get compute that does not require Nvidia. But they are still constrained by chip production at TSMC. Now how is this a disadvantage well most researchers prefer working with CUDA which means most training of AI models is done on Nvidia GPUs. This means that if they want to do inference on TPUs, there needs to be changes which can reduce the models performance.

What X AI has

The recent acquisition of Cursor by X AI has really changed the game as Cursor has good strengths where X AI had weaknesses. Let's first talk about researchers while X AI does not have that many prominent researchers they still have a good talent pool. Cursor seems to have had one of the best teams for reinforcement learning as seen with Composer 2.5 where they did reinforcement learning on Kimi K2.5 to massively increase the coding performance.⁵

Now let's talk about compute this is where X AI is doing very well as they have built two massive data centers called Colossus. In a normal market where compute was less constrained the amount X AI built would have been a bad idea, but due to the difficulties of getting GPUs from Nvidia it's been very successful.⁶ If the supply for power and Nvidia GPU shortage and the lack of usage of X AI models building so much capacity would have been a bad idea. But you can see with the deals with Google⁷ and Anthropic.⁸ Both of these deal make up around 84% of X AI's Revenue according to the S-1 filings.⁹

Data is where the recent acquisition of Cursor makes a whole lot of sense as X AI has good General data from Twitter/X, But did not have that much code data. Cursor has this type of data which X AI is lacking because much like Google Cursor Collects users requests and is allowed to use them for training new models. This type of data is the type of data X AI is lacking.

The recent X AI's recent acquisition of Cursor is already paying off. As seen with the release of Grok 4.5 which put it in the third position on artificial analysis intelligence benchmark, until today with the release of Kimi K3 which has displaced it for third place.¹⁰

The future

The future of Google and DeepMind is really riding on Gemini Pro 3.5, as if it's only slightly better than Flash they are going to really struggle to gain market share. If Gemini 3.5 pro has been in the oven longer for a good reason as is a Sol/Fable class model it's going to put Google near the top again.

For X AI the future looks bright with the full acquisition of Cursor and the potential of a Grok 5 which has both teams working alongside each other, they can keep their position in the top three AI players. If they do what Elon said and release a new model every month this will allow them to refine and make incremental improvements on models which should keep them near the top of the charts.

DeepSWE. (2026). DeepSWE. https://deepswe.datacurve.ai/ ↩︎
Eldridge, A. (2026, May 11). Google DeepMind | History, Innovations, & Controversies. Encyclopedia Britannica. https://www.britannica.com/topic/google-deepmind ↩︎
Reuters. (2026, June 19). John Jumper to leave Google DeepMind for Anthropic. CNBC. https://www.cnbc.com/2026/06/19/john-jumper-to-leave-google-deepmind-for-anthropic.html ↩︎
Sen, M. (2026, March 25). Google Gemini Statistics 2026: Users, Revenue & Growth. Blogs | Panto AI. https://www.getpanto.ai/blog/google-gemini-statistics ↩︎
Cursor Team. (2026, May 18). Introducing Composer 2.5. Cursor. https://cursor.com/blog/composer-2-5 ↩︎
Colossus | xAI. (2024). X.Ai. https://x.ai/colossus ↩︎
Kolodny, L. (2026, June 5). Google to pay SpaceX $920 million a month for compute capacity at xAI data centers. CNBC. https://www.cnbc.com/2026/06/05/google-to-pay-spacex-920-million-a-month-for-xai-compute-capacity.html ↩︎
xAI. (2026, May 6). New Compute Partnership with Anthropic. X.Ai; xAI. https://x.ai/news/anthropic-compute-partnership ↩︎
Space Exploration Technologies - S-1. (2026). Sec.Gov. https://www.sec.gov/Archives/edgar/data/1181412/000162828026036936/spaceexplorationtechnologi.htm ↩︎
Artificial Analysis. (2026). Intelligence. Artificialanalysis.Ai. https://artificialanalysis.ai/?intelligence=artificial-analysis-intelligence-index ↩︎

Will The Steam Frame Lead to a Wave of Arm Linux Laptops

Micah Wade — Tue, 14 Jul 2026 00:00:00 +0000

What Valve Has Been Working on

While you might not relate to this my friends on our discord server keep thinking the Steam Frame is going to release tomorrow for the last week. While Everyone is hoping that the Steam Frame is at a competitive price. My hope is that this kick starts a wave of Linux laptops

The Best thing valve has been working on along side the open source community is FEX-Emu. FEX-Emu allows for x86 programs to run on Arm via emulation.

The State of Arm on Linux

While the Linux kernel has had support for ARM since 2002 app and distro support for it has been up and down. While arm has gained popularity in servers, and a bit of growth in laptops with windows on Arm. The support of apps on Arm has been the major issues because while in a server you would only run 5 applications max. Personal computers would have 20-100+ applications installed, much like people switching from Windows if one app does not work that can be a deal breaker.

The big layer with Linux ARM support is Raspberry Pi. Which due to being the controller of the software and hardware, similar to Apple they have a more curated experience. While Raspberry Pi's are amazing they don't have the best performance compared to what you get in other ARM chips.

What Would Run Natively

For my main computer I am running fedora with KDE plasma and try to install everything from the Discover store. To test what percentage of apps would work natively run on ARM. I told Gemini to, "Run though everything I have installed in the Discover store and to tell me what percentage would run natively or ARM."

In the end I had 122 GUI apps installed and 116 had ARM support. This was higher than I was expecting, but if one of the apps you need does not support ARM it's a big deal breaker.

If you want to check your go to this repo: https://github.com/MicahWade/discover-arm-checker

Perfect ARM Discover Store

As a user of KDE with Fedora the Discover store is how I install everything. It's also the reason if someone asks more for a distro choice I recommend Fedora with KDE because everything is so easy to install.

Now that we know why the Discover Store is so good, how can we make it even better on ARM? To start with if you are installing it on ARM CPU the first time you open the store, the store could ask you if it should use FEX-Emu for apps that don't have ARM support. This means whenever you are searching for something it should show up as an option. Because this would be for less than 10% of apps the small performance hit will be negligible. This would mean that it feels like any other Linux computer.

Future Thought

With the most likely success of the Steam Frame I see there being more arm laptops coming to market. While they may not be Linux the support and money Microsoft is putting into windows on ARM will make companies more comfortable developing ARM laptop.

While there have been some Linux ARM laptops like the Argon ONE it uses a Raspberry Pi CM5 computer board. While this is good it's not at the performance levels most people expect from a laptop. Another option for a Linux ARM laptop is putting Linux on a Windows on ARM laptop. The current issue is while there has been great work adding compatibility for the snapdragon chips, it's not ready for day to day work yet.

If you want a Linux ARM laptop and a CM5 does not give the enough performance, the best option in my opinion is the ARM Framework main board made by MetaComputing. But this board is not officially supported by Framework and is less efficient than other board options by Framework. But if you need ARM because of the work you do on your laptop, it's the best option. In 1-3 years I can easily see Framework having an official board that has an ARM processor, this I could see most likely coming to the Framework 12 first.

While this is great there is always RISC-V to consider as in a few years it could be competitive to ARM. If RISC-V is even just competitive with ARM not having to pay licensing cost could see it become the king.

$100M For Canberra to Sydney Train

Micah Wade — Mon, 13 Jul 2026 00:00:00 +0000

Current Issues With The Canberra Branch

There have been upgrades on the Canberra Branch that have been long overdue as stated by the government.¹ As someone who rides the train semi regularly, I 100% agree with as you can feel the tracks a lot more there compared to when you get past Goulburn. One Big issues with the line is speed as it goes through the Molonglo Gorge.² It's also slow because the line is old and wasn't designed for the speed of trains today.

One big issue which is not 100% the tracks fault is frequency due to the popularity of the lunch time train service it's almost always booked out. Which is where you think why don't they just run more trains? The current issues is that the rolling stock the Xplorers, are limited and were built in 1993 so even if you wanted to add more train there just are not enough in the network. Now this issue is being fixed by the new rolling stock the R-sets³ which while looking good, I don't think they are getting the amount that will be needed.

Now where does the track come into this? Well if they want to run trains close to each other because people want to get the train around lunch time let's say a service leaving Canberra at 1100, 1200, 1300. That would not be able to be done with the current tracks even if we had the trains lots of the it is single tack and you would have the morning train from Sydney coming the other way.

Why Now is a Good time?

With the new train sets coming into service maybe late 2027 it's not very clear. There will be plenty of time to do the required track work before the new trains go into service. This hopefully will allow for faster speeds on parts of the track and not track closures when the new trains enter service. I think it's critical that when the new trains enter service there is the least amount of interruptions possible due to people who may have not considered the train considering it because of the new trains.

Target Journey Time

With the new track works happening and the New train fleet coming into service I have been thinking about what the Journey time needs to be in order to be competitive. To start we need to look at the two bigger competitors driving and catching the bus. Let's start with driving if go from Canberra to Sydney Central the time it takes is 3hr 13min with tolls.⁴ Now let's look at the bus option I looked at getting for Canberra to Sydney central according their website it will take about 3.5h.⁵ Now if we look at transport NSW it time is 4h and 8 minutes with it being a few minutes late most times I take it.⁶

So taking this data let's look at what time we should be aiming for. While the driving option looks like the best option for speed I don't think most people are doing that drive without a stop. If we add just a 10 minute stop which is not the shortest but also not huge it's around the 3.5h mark. Considering that I think the train should be aiming for a speed of about 3.5h to Sydney which I don't think that just this track work can meet but if they do more work in future and potentially electrification up to Goulburn. This is a reasonable speed to achieve without spending billions on High speed rail.

Other improvements

Even though other options may be faster even after the track work the train still provides the best comfort. Compared to driving it has a huge bonus if you're the one driving, which I will talk more about in another post. The ability to do other things you don't have to focus on the road you can read a book, play games or watch a show. But it also has advantages over the bus you can get up and stretch your legs got to the buffet car and get something to eat and drink.

Now that I have talked about the advantages of the train, let me tell you about why it's worse compared to other options currently. The big ones is frequency 3 options a day. Compare this to the bus which runs every hour or the car ride were you can leave when ever. With the new train this should be fixed and they run more trains but we don't know yet.

Location Jolimont centre has the advantage of being right next to the city bus interchange, Compared to Canberra station which has one bus line going to it from the city. Now how do we fix this in a perfect world we would have trains go into the city but that ship has sailed. But what we can do is a few things first have a better bus shelter there with added signs, which make it easier for people who have never been to Canberra understand what they have do. This should mean there are posts telling people they can pay with there cards, but also signs telling them where to go and which bus to catch depending on where they want to go. Second there should be more busses from different interchanges which busses arriving for every train frequency from Woden, Tuggeranong, Belconnen and Weston Creek not just the city so most trips can be done with a single bus change. Compared to the 2 changes from most places.

Final Thoughts

While this is great that they are doing these upgrades in my opinion as they are long overdue. But if they want to increase ridership which will need to happen if we want to build a low carbon future they should think about other changes like people connecting to the train in Canberra and the frequency. If some money goes into these parts I think it will drastically increase ridership along with the new trains and these track works.

ACT Government. (2026, May 12). Funding boost to improve Sydney to Canberra rail link. Our Canberra. https://www.act.gov.au/our-canberra/latest-news/2026/may/funding-boost-to-improve-sydney-to-canberra-rail-link ↩︎
History of the Canberra Branch. (2026). Canberrarail.Org.Au. https://www.canberrarail.org.au/historycbrbranch ↩︎
Transport for NSW. (2017, May 31). Regional Rail. Www.Transport.Nsw.Gov.Au. https://www.transport.nsw.gov.au/projects/current-projects/regional-rail ↩︎
Google. (n.d.). Driving directions from , ACT to , Haymarket, NSW [Map]. Google Maps. Retrieved July 13, 2026, from https://maps.app.goo.gl/o3qHuYyLRpaap9Cz8 ↩︎
Murrays Coaches. (n.d.). Express tickets booking: Canberra to Sydney Central. https://online.murrays.com.au/murrayscoaches/booking/expresstickets ↩︎
Transport for NSW. (n.d.). Trip Planner results [Interactive route results]. https://transportnsw.info/trip-planner/results/aa9a56a1c28f178ee65b89c4d7055d61?excludedModes=11&from=26041&to=200060 ↩︎

How to Setup Ubuntu Minimal With the Lowest Resource on Your Server

Micah Wade — Mon, 13 Jul 2026 00:00:00 +0000

What is Minimal Compute

This guide is aimed at servers with minimal compute for this i consider anything with less than 2 cores or less than 4GB of ram. For the guide I will be using the free compute Oracle gives you which is 1/8th of a CPU and 1GB of memory. If you are running this on your own hardware I would recommend installing Alpine Linux instead. The reason I made this guide is for people who are running their server in the cloud because they may not have the option of using Alpine. This is what I did when setting up the Ubuntu Server that this is running on.

The Two Constraints

So when we are optimising there are two main constraints: one is CPU it can only do so many instruction, two would be memory there is only so much which can be stored.

Memory

Swap

When dealing with minimal memory we want to make sure the server does not crash, because it ran out of memory even if that means it's slower. To alleviate this we will create a swap, which means when the memory is close to full it will move the least important things onto the disk. While the disk is slower it can save the server from crashing. Due to us only creating 2GB and we are doing this with a cloud VPS the Drive it uses will be plenty fast the main issues with be latency.

First we will reserve 2G for the swap:

sudo fallocate -l 2G /swapfile

Then will make sure it has the read/write permissions:

sudo chmod 600 /swapfile

Next we Need to tell the Linux kernel that this file is for virtual memory:

sudo mkswap /swapfile

Next we need to activate the swap file:

sudo swapon /swapfile

In order to confirm this worked you can run:

free -h

If it worked you should have something like this:

              total used free
Mem: 952Mi 188Mi 137Mi
Swap: 2.0Gi 23Mi 2.0Gi

Great now this is working but it won't stick around if we reboot. In order to fix that issue we need to mount this swap file in our fstab:

echo '/swapfile none swap sw 0 0' | sudo tee -a /etc/fstab

Snap

Snap is a package manger mostly only used by Ubuntu most servers don't use snap. If you don't intend to use snap you can remove it to get a small memory reduction.

To start we want to remove any snaps installed:

snap list | awk 'NR>1{print $1}' | xargs -I {} sudo snap remove {}

Now that everything installed through snap is gone you can remove it through apt:

sudo apt purge -y snapd

That should have got rid of everything but snap has some dependences which may not get removed. To remove these we will scan for the packages that were installed as a dependency but are no longer needed:

sudo apt autoremove --purge -y

Well done you have now deleted snap.

If you ever want to use snap you can install it with:

sudo apt update && sudo apt install -y snapd

CPU

Disable Daily Upgrades

While Updating and Upgrading in the background can be good, if your server is already working to the max it can cause issues when you have little CPU power. But be careful if this is a public facing server i would recommend you setup a script to do updates when it has low CPU usage as this can be a security risk.

sudo systemctl disable --now unattended-upgrades.service
sudo systemctl disable --now apt-daily.timer
sudo systemctl disable --now apt-daily-upgrade.timer

If you Disable these make sure to be running updates and upgrades regularly:

sudo apt update && sudo apt upgrade -y

AppArmor

AppArmor is great and you should probably not disable it. But if your trying to get every tiny bit of performance and make sure all your other security is high it's most likely okay.

To disable AppArmor run this:

sudo systemctl disable --now apparmor

Conclusion

Now as i stated above you should probably be using Alpine Linux. But if you insist on using Ubuntu this how you can need to reduce the base line system usage of Ubuntu.