Jacob Ortiz 🇺🇦

[Jacob Ortiz 🇺🇦 @Hip_Chimp]

@benjamin_horne Can be used to review the generated document to verify compliance. This can allow a person to massively accelerate their productivity. I do not think that AI will replace people, but it will make people's jobs easier.

[Jacob Ortiz 🇺🇦 @Hip_Chimp]

@benjamin_horne I think this is a bad assessment. A lot of white collar work is referencing rules to create compliant documents, and a properly harnessed and managed AI agent can complete a significant amount of work that high quality in little time. The time you save drafting the document. . .

[Jacob Ortiz 🇺🇦 @Hip_Chimp]

@josusanmartin I've had a lot of success by basically building a four man team consisting of myself, an AI PM, an agent that drafts what are essentially Jira tickets, and a coding agent that does the actual implementation. It's worked fairly well so far. I'm in the process of formalizing it.

[Jacob Ortiz 🇺🇦 @Hip_Chimp]

@PlsHoldMyHalo @d29756183 This is a poor assessment. Very powerful models are open source and can be run on relatively minor hardware. You can run DeepSeek V4 Flash on a single DGX Spark. The reality is that you can already do most of this if you have a bit of knowledge and AI assistance.

[Jacob Ortiz 🇺🇦 @Hip_Chimp]

@ShriramKMurthi AI models can't see letters, they think and generate in tokens, which don't have specific numbers of letters. Asking an AI to count letters is like asking a colorblind person to tell you if a traffic light is green or red.

[Jacob Ortiz 🇺🇦 @Hip_Chimp]

@AguiadosFactosV @AltSamCat 1492 was a long time ago.

[Jacob Ortiz 🇺🇦 @Hip_Chimp]

@FrankRundatz @ewanm Token prices have crashed, and you can save a lot of tokens by optimizing workflows with detailed scaffolding and intelligent context management. You can work with frontier models for the heavy planning and architecture, then task a less expensive agent with the grunt work.

[Jacob Ortiz 🇺🇦 @Hip_Chimp]

@seanlinehan @ameliakadams I switched off Claude because of the insane usage limits. I used to work with it all the time, but Chat is there when I need it, especially for research and project planning, which draw from the same bucket as agentic applications on Claude.

[Jacob Ortiz 🇺🇦 @Hip_Chimp]

@MJKobeFlocka @JeremiahDJohns You can be weak, or you can be a leader, but you cannot be both.

[Jacob Ortiz 🇺🇦 @Hip_Chimp]

@per_arneng @DaveShapi Being "satisfied with your job" is not the same as "getting paid to do what you love". I'm relatively satisfied with my job, but I don't get paid a lot and it's not what I love. The position*you* are in, getting paid a lot to do what you love, is an impossible dream for most.

[Jacob Ortiz 🇺🇦 @Hip_Chimp]

@per_arneng @DaveShapi You don't see, at all, how your position is different from 99.99999% of people? Good for you that you get paid a lot to do something you love, most people get paid not enough money to do something they hate. It's probably best for you not to further share your ignorance.

[Jacob Ortiz 🇺🇦 @Hip_Chimp]

@jmbollenbacher @perrymetzger AI doesn't do that stuff, people do. AI can't choose to do much of anything. It's probably easier to hate AI than deal with people, but it is very difficult for AI to do harm independently.

[Jacob Ortiz 🇺🇦 @Hip_Chimp]

@Gerrylwk @j_dekoninck @thkostolansky Yes, you're missing everything. LLMs are probalisitic generative platforms, they are completely incapable of producing deterministic output.

[Jacob Ortiz 🇺🇦 @Hip_Chimp]

@DerekPederson3 @glencoe2004 @Tess581670 @PreferNoHandle @Noname117S You wouldn't power a starship with a rocket, you'd use an ion drive or something. In any case it would probably be more realistic to figure out a warp drive or some other way of avoiding acceleration of matter.

[Jacob Ortiz 🇺🇦 @Hip_Chimp]

@staysaasy Why are you coding with prompts?

[Jacob Ortiz 🇺🇦 @Hip_Chimp]

@ProphetKG @Hedgeye The price is going to crash as soon as the IPO is over. We're getting fucked.

[Jacob Ortiz 🇺🇦 @Hip_Chimp]

@teortaxesTex This isn't stupidity, it's bad context management. The LLM doesn't know why you want to go to the car wash, and you poisoned the context by describing the car wash as "only" 100m away. If your question was more clear you would get the right answer.

[Jacob Ortiz 🇺🇦 @Hip_Chimp]

@codeRunnerUK @gailcweiner That's not how refusal works. Refusal works by having weights in the model shaped by examples of questions where the correct answer is to refuse to assist, but these activations take place in the high dimensional matrix, not the reasoning chain. The CoT is just a verbalization.

[Jacob Ortiz 🇺🇦 @Hip_Chimp]

@svkenney AI is not code. LLMs are not authored, they are trained, essentially grown. That's why the output cannot be predicted or controlled beyond a certain amount of influence.

[Jacob Ortiz 🇺🇦 @Hip_Chimp]

@jere_codes @TiernanDeFranco What I'm saying is that it's pointless gatekeeping. It's the same as with Billy West. He doesn't value his skills intrinsically, he only values them as a status symbol he can hold over others.