Ben T. @idontnull

Founder & CEO of https://t.co/vb2XQ0HQhc itva.com Joined May 2024

Tweets

37
Followers

28
Following

53
Likes

9

Ben T. @idontnull

a year ago

Nothing makes me less bullish on AI and LLMs than trying to get it to do regex.

0 0 1 18 0

View Details

@JBizzle703 You could make leetcode for network engineers by spinning up virtual sandboxes of IOSv, VEOS, etc. It would just be much more resource intensive than leetcode which just needs to execute code in a sandbox. But with an actual virtual environment you can run true validation.

0 0 1 19 0

View Details

Ben T. @idontnull

a year ago

@Lyle_AI @garrytan Hard agree with you, true customer adoption is when they don't even care how the technology works. When we finally got our first couple Fortune 500 customers they never asked how the product worked. Only asked if it can get "x,y, and z" results for them.

0 0 1 60 0

View Details

Ben T. @idontnull

2 years ago

@alliekmiller Agreed, distribution is more important than the product. AI Adoption is slow and noisy. People are "AI Fatigued" with everything being labeled with AI. The moment we removed AI from our messaging and just focused on the benefits, we saw huge increase in adoption.

0 0 0 62 0

View Details

Ben T. @idontnull

2 years ago

@DataBunnyLLC Most people starting out, even those with CCNA, are always lacking in Firewall. I tell them buy a cheap used Fortinet on ebay. Start with basics and replace your home router with a Fortinet FW. Then play around from there(e.g Filtering, SSL VPN, NAT, IPSec Tunnel, etc.)

3 4 76 3K 50

View Details

Ben T. @idontnull

2 years ago

@cloudtoad Probably low. Can't blame the network if they can't even boot up their computers to message the networking team.

0 0 2 108 0

View Details

Ben T. @idontnull

2 years ago

@brianwhelton Definitely a it depends! I should have specified I prefer Friday nights. At least before Monday start you can let people know there are issues. If I got on a plane right now from LA I still wouldn't make it in time, hopefully next time! Hope you all have fun!

0 0 1 15 0

View Details

Ben T. @idontnull

2 years ago

@AdoraNwodo With outages that size, even if you're not on call, you're getting called. Outages or issues that size tend to result in a 300+ person slack/teams group triage

1 2 11 3K 1

View Details

Ben T. @idontnull

2 years ago

@alexgraveley Whoever builds the product where my mom doesn't even realize she's using AI. It just happens in the background and she is happy with the results.

0 0 0 57 0

View Details

Ben T. @idontnull

2 years ago

@abacaj Synthetic data has to be used to some extent. All the data we need to build AI Agents isn't out on the internet or anywhere at all. When engineers solve problems they don't write out all their individual steps and thought process and post it online. You need to make that data.

0 0 1 70 0

View Details

Ben T. @idontnull

2 years ago

@emollick This is my favorite part of this LLM boom, there's no "I'll never catch up". It's still early. It's also incredibly fun and unique opportunity to DISCOVER. You get to discover new things instead of repeating best practices.

0 0 2 259 0

View Details

Ben T. @idontnull

2 years ago

When you troubleshoot a networking issue with no details, what order you go in? I always do Layer 3 -> Layer 2 -> Layer 1 -> Layer 4 -> Layer 7

0 0 0 188 1

View Details

Ben T. @idontnull

2 years ago

Qualified Network Engineers are so difficult to find that even Elon has to personally advertise open roles

Elon Musk @elonmusk

2 years ago

Both @Tesla and @xAI are looking to hire networking engineers & technicians

8K 13K 157K 56.7M 4K

0 0 0 253 0

View Details

Ben T. @idontnull

2 years ago

Going to be huge for Enterprises and less so hobbyists. A lot of Enterprises are waiting for capable models that can be air gapped. The compute cost to run a 400B size model is nothing for them. Interested to see if they go compute resource per AI app or shared compute infrastructure amongst all AI apps.

0 0 0 160 0

View Details

Ben T. @idontnull

2 years ago

Agreed, the hardest part is getting the first job. Job offers don't magically come after getting a cert. Do anything to stand out. There are so many people I've seen when hiring for entry level that get a cert but don't even touch technology. Buy a used a switch on ebay, play around with linux, etc. ANYTHING to separate you from the pile of hundreds of resumes that have just the same cert as you with no experience. Cause you are competing against people with experience and personal projects.

0 0 3 371 2

View Details

Ben T. @idontnull

2 years ago

@TriniNetEng I've had so many companies tell me the senior engineers are the biggest bottleneck for very similar reasons

0 0 2 72 0

View Details

Ben T. @idontnull

2 years ago

@bindureddy Fine tuning doesn't even add data. Even if it did, the chances of you having LLM-optimized data no one else has is slim. Training from scratch is so expensive if you do. Fine tuning barely beats in-context examples in the leading models.

0 0 0 28 0

View Details

Ben T. @idontnull

2 years ago

@JvNixon The most insane part is a 55/45 Consumer/Enterprise split. I'm certain the Enterprise split will far outgrow the consumer as more and more applications are built for specific workflows instead of a catch-all chatbot.

0 0 2 2K 0

View Details

Ben T. @idontnull

2 years ago

The lack of LLMs willingness to admit they’re wrong is just a representation that we never really admit we’re wrong. Especially on the internet where the training data comes from.