cole murray
@_colemurray
Followers
2K
Following
11K
Media
2K
Statuses
5K
ai/ml | cto | second time founder | former sr. sde @ amazon
San Diego, CA
Joined February 2015
Advice given to someone asking about AI Consulting:. I don't think an ML background is required to be successful in AI consulting, but obviously helps. I think the biggest "skill" learned in ML is how to successfully do feedback loops in a system. In an ML system, this typically.
6
15
171
psyop successful . we got em boiz . - developer productivity absolutely tanked.- hiring ability tanked. well done all around.
3
1
28
I used to be an ftp hater until I landed a 7 figure deal for a client with a legacy corpo. I hate no more . emailing passwords, shared creds, no audit logs. all in a days work.
1
0
7
felt this claude opus degradation over the past few days. mildly relieving to see the status confirmation
7
13
102
when they came for the entire internet via CDN globalization I said nothing. when they came for my portfolio, I was outraged. -. The argument being made is that it is concerning that cloudflare will be the defacto registry of “AI trust”. brother… they are already the internet
0
0
3
pretty sure this was just a call from Jensen telling the devs to do something about NVDA being down post earnings.
OpenAI appears to have announced some kind of breakthrough internally yesterday, and the entire staff were given license to vaguepost. Let's hope it doesn't take too long for us to find out what it was.
0
0
1
i use to do this at Amazon to look for bad product recs. find bad reviews -> identify root cause driving it -> fix.
I just listened to a breakdown of Gemini jump from 2.0 to 2.5, and here's the gist. The team did not rely on lab tests alone. They scraped real user feedback from X, turned those "this broke" moments into living evals, and keep appending new edge cases with every release. It's
0
0
0
small AI teams get deceived into relying on explicit customer feedback . thumbs up / rating doesn’t scale and isn’t representative of your systems behavior. your feedback is the implicit user behavior.
0
1
4
did you guys create this in-house for clicks?. no serious actor would do this, let alone with a 20b model.
#ESETResearch has discovered the first known AI-powered ransomware, which we named #PromptLock. The PromptLock malware uses the gpt-oss:20b model from OpenAI locally via the Ollama API to generate malicious Lua scripts on the fly, which it then executes 1/6
0
0
4
we had chrome LLMs before it was cool.
I'm tired of software that works against me and allowing 3rd party devs full access to my browser for tasks. In a 1-hour session with Claude (I timed it lol), I built an extension for summarizing text I highlight. (code below)
0
0
2
evals show you known failures. observability shows you unknown failures. one is more important than the other.
0
0
3
reading the docs and occasionally the source will make you better than 90% of your peers.
0
0
3
should I use ABC platform for my agent?. given you are asking, you're probably in the build stage and not the optimize stage, so go for it. find what works and optimize later.
1
0
5
There’s vulnerabilities and then there are “vulnerabilities”
It's 2025. There is no excuse for any product to past tokens into SQL statements like this, much less a security product. Fortinet, a cybersecurity company, has become famous over the last several years for its security flaws. An occasional flaw is forgiveable, but this line of
0
0
2
RT @GeoffreyHuntley: retweets appreciated. hi folks, some important life news.i’m looking for a new employer - contact via ghuntley@ghuntle….
0
103
0
why can't i just go to github and see all the security issues with a repo. i come back to this question a lot.
0
0
3