馃Startup founder, doing something great!馃グ Open Source: https://xitter.cf/t.co/VD62jUqDhl My GitHub: https://xitter.cf/t.co/waq3Lg4Cw4 I always follow back!
Joined October 2022
- Tweets3.6K
- Following1K
- Followers911
- Likes3K
JustJerry retweeted
Overheard: MLE, as Machine Learning Engineers(not researchers) will be the next job being replaced by AI, faster than senior engineers.
Turns out, AI is very good at testing parameters to tune and optimize the shit out of it, given a clear optimization goal.
Agree?
JustJerry retweeted
la plupart des gens 茅changent leur audace contre l鈥檃pprobation sociale vers 25 ans et passent les 60 ann茅es suivantes 脿 regretter ce trade
脿 8 ans vous vouliez conqu茅rir Mars et 脿 28 ans vous h茅sitez 3 heures avant de publier un tweet de peur de manquer de s茅rieux et d鈥櫭猼re moqu茅
entre les 2, vous avez int茅gr茅 la crainte du jugement de gens qui vous auront oubli茅 quelques semaines voire quelques mois apr猫s votre mort
l'erreur la plus tragique consiste 脿 atteindre 80 ans en r茅alisant qu'on a pass茅 plusieurs d茅cennies 脿 solliciter la permission d'exister juste parce qu鈥檕n 茅tait p茅trifi茅 par le regard d'autrui
There is no need for FOMO. Just think about the people who haven't even been born yet: they are missing out on this current AI explosion, but there will always be opportunities unique to their own generation
Lately, the frequent resetting of Codex usage limits has given me a slight sense of unease regarding tech stocks.
If companies like OpenAI and Anthropic stop their aggressive purchasing of computing power, then the valuations of these tech stocks simply won't hold up.
If a product offers a user experience that goes above and beyond Notion, while allowing you to power its internal AI features using your existing subscriptions鈥攍ike Codex, Claude Code, or other Agents鈥攚ould you use it? Most importantly, it is open-source and your data stays entirely local. Come give BaseHalf a try; contributors are more than welcome.
AWS Activate credits finally came through!馃グ
the process was less about one magic fix and more about getting the boring details right
if you鈥檙e stuck in the rejection loop, happy to help
The strong should treat expectations as a tool.
To accomplish great things, we must have high expectations, as they provide the dreams and motivation necessary. However, when it comes to the actual execution, we must not let our expectations remain too high, as that leads to suffering.
Therefore, the strong should use expectations as an adjustable tool: set them high when inspiration is needed, and lower them when it is time to work steadily and focus on the tasks at hand today.
For people like us, success is only a matter of time. We internalize every periodic setback and failure, turning them into a continuous stream of motivation to move forward. Never ever give up馃
JustJerry retweeted
We can basically think of an agent as a GitHub repo.
The LLM is an executable inside it. The context is a set of text files. The harness is the code that manages how context and the LLM interact.
Apply harness engineering to this repo, and you get a self-improving agent.
To all startup founder on their journey:
I understand that the technological democratization brought by AI can be disorienting, especially after experiencing what tools like Fable can do.
However, remember that software is ultimately a means to satisfy user needs. Software itself holds inherent value; I personally continue to pay for various services, not just Codex and Claude. For those in the software or gaming industries, maintaining your mental fortitude is now even more critical than strategic planning. The priority is simply to stay in the game. As long as we remain at the table, our odds of success increase exponentially. Software matters, but the real challenge is the struggle within ourselves. We must remain steadfast in our conviction that we can build something meaningful and create genuine value for society.
Fear and uncertainty are natural responses. Yet, if we commit to showing up every day, staying focused on the present, and striving for incremental progress, we maintain an upward trajectory.
If you persist with this mindset, you will undoubtedly build something great.
JustJerry retweeted
kinda crazy that someone's full-time job was to steer claude to sabotage ML research capabilities for paying customers
This is a real phenomenon, and I鈥檝e actually encountered this issue in practice. This Skill filters out certain words鈥攆or instance, removing noise like "help me," "I think," and so on. Truly valuable data definitely needs to undergo data cleaning; it consists of one's own perspectives and reflections, rather than just a collection of imperative commands. Large language models can help us extract our ways of thinking and underlying personality traits from this data, which is actually quite interesting
Your unguarded data beating a quiz you know you're taking is a real insight. The thing I'd watch: a lot of what's in your prompt history is you at work solving tasks, which is a different person than you off the clock. You might read someone as cold and systematic when that's just their job leaking into the logs. How are you separating the work self from the actual person in the denoising?
who-am-i skill launched! 馃殌
This is a skill that uses your local data to analyze your personality. Those multiple-choice tests are conducted consciously and may not be accurate. However, when you generate this local data, you aren鈥檛 aware at all that it will be used for personality analysis, making it highly effective.
The data includes everything stored locally, the prompts you鈥檝e entered, and your Typeless history. After the data is denoised, this skill will tell you what your personality is like. Trust me, it will help you discover a side of yourself you never knew.
All data stays local, so there is no need to worry about privacy馃槈