Philip Kung
@PhilipKung5
Followers
632
Following
675
Media
67
Statuses
632
building @SweetspotGov (YC S23), formerly SWE at @meta , @nuro , @amazon , @coinbase
New York City
Joined June 2020
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Lakers
• 140528 Tweets
Fernando Valenzuela
• 117376 Tweets
LINGORM FILL
• 82975 Tweets
期日前投票
• 82634 Tweets
#MerzAestheticsXหลิงออม
• 79158 Tweets
メニュー
• 65510 Tweets
助成金2000万円
• 64819 Tweets
非公認議員
• 61719 Tweets
Kürt
• 61066 Tweets
非公認候補の支部
• 46610 Tweets
प्रियंका गांधी
• 31907 Tweets
DONNY 4everAKapamilya
• 31650 Tweets
非公認の候補者
• 31544 Tweets
Daisuke
• 25277 Tweets
ILL BE THERE POSTER
• 18340 Tweets
党勢拡大のため
• 17273 Tweets
PRABOWO-GIBRAN KerjaFOKUS
• 15342 Tweets
SiapBEKERJA UntukRAKYAT
• 14931 Tweets
APT 100M VIEWS
• 14333 Tweets
Müsavat Dervişoğlu
• 14162 Tweets
#わたしの推し牛
• 13755 Tweets
キューティーハニー
• 12415 Tweets
気圧のせい
• 11857 Tweets
@iamyesyouareno
i love how they try and make him look innocent with the glasses 😂, oh lawyers
12
7
1K
@napoleon21st
same thing with my friend too, they stuck a monkey on his shoulder at a flea market and wouldn't take it off until he paid a tip 😂
7
10
363
@tsarnick
these universities have endowments larger than the cash reserves of most tech companies. let’s not pretend that they can’t afford to get the resources necessary to train competitive models
10
4
224
@NotTuxedoSam
im guessing its because its all read no write since (i assume) < 0.1% of the traffic actually makes an account
3
0
126
Pleasure meeting AirBnb CEO
@bchesky
at the
@ycombinator
office in Mountain View today! Amazing talk - was inspired by how Airbnb sold cereal in their early days to fund their operations.
8
9
82
@AiBreakfast
why would they? apple is not in the business of enterprise software so it doesn’t matter that they don’t have a frontier model.
apple had the foresight to create custom silicon that can run these models and deploy it three years back. now they can take advantage of all the
11
3
53
@teortaxesTex
this is very true. to an extent, chinas current rise against the west has been a 200 year revenge plan to atone for the humiliation suffered under the opium wars
1
3
49
nice, very subtle OpenAI
OpenAI are working on a reasoning technology that they are now calling Strawberry and which was previously known as Q*
148
221
1K
3
3
46
@meetbarvadiya
@tsarnick
why is it the goal of a university to prioritize making a profit from a stock portfolio rather than maximizing technical and educational impact to society?
1
0
19
the latency of the new gpt-4o model is insane - its running at close to 180 - 200 tokens per second for text output. time to first token is also near instantaneous with thousands of input tokens. openai did an amazing job - congrats to the team.
2
3
18
@PatrickJBlum
our friends on cost plus contracts could never. congrats to anduril for being ahead of schedule on this.
0
0
18
first day at sweetspot, pushed a little update and taking the day off
YC S23's
@SweetspotGov
has added 200,000+ state, local, and education contracts to their platform — allowing users to leverage their AI tools to win even more bids.
Congrats on the launch,
@sachsubra
,
@PhilipKung5
, and Andrew!
3
18
81
1
1
14
almost one year since we've gotten our YC acceptance - at this time
@sachsubra
and i were still building our food ordering startup out of his living room. damn time flies.
1
1
13
My first marathon! Signed up impulsively last week w no training but am super satisfied with my performance.
I underestimated the difficulty of running a marathon and overestimated my muscle endurance for sure 😂
3
0
12
bad day to be a cigar
1
0
12
1
3
11
watching qdrant and weaviate employees beef across github, x, and hacker news is pretty funny
1
0
11
2
0
10
@abacaj
agreed, its similar to training models where the underlying data quality is more important than whatever hyperparameters you set.
1
0
10
dev tools should have a copy of their documentation stored in an llm-docs.txt file to make it easy to inject that info into a coding agent
4
1
10
@IIDrxpzy
@veroniccasdad
@jakepaul
religion is useful for society because it provides a baseline code of conduct. most religions follow similar principles - love thy neighbor, don’t commit adultery, etc. ofc some people take it too far but without religion, society descends into hedonistic madness as we see now.
2
0
8
@emollick
lets wait for the new data centers to turn on first. new frontier models require new chips
1
0
10
no one has any reaction to cursor raising a series A at a 400 million dollar valuation on X??
5
0
8
@bindureddy
openai needs to deprecate 3.5 turbo ASAP - so many people rely on it to do batch jobs and data enrichments. switch it to an alternative that is half the size and twice as fast and move that compute over to gpt-5 🙏
5
0
9
@_xjdr
im guessing fireworks has a combo of crazy speculative decoding framework + AMD hardware that allows them to serve at that price - otherwise its the VC subsidy
0
0
9
i literally get faster internet in the mountains of vietnam than in the heart of SF. fucking embarrassing 😂
0
0
8
@raghavar0
@KrutrimAI
Even Grok and Mistral had this issue - doubt it’s a wrapper around OpenAI
1
1
9
@eterevsky
@tegmark
seems more like a hardware issue - if you modeled this scenario in a simulation i'd bet money on gpt-4o being able to solve it.
5
0
8
Lex Friedman is dangerous - decided to put on the Elon podcast while driving and nearly dozed off and crashed. my girlfriend had to switch over while i recovered from the drowsiness.
i would pay so much money for someone to dub him over with an eleven labs voice.
1
0
8
Met
@commadangle
and
@thewizardlucas
from the
@GetErgomake
team at the YC 23 retreat. Amazing product - heavily recommend!
0
1
7
@mattshumer_
@SupermavenAI
I tried
@SupermavenAI
and had to uninstall unfortunately. I like how concept of the large context window but unfortunately the reasoning is not that good, leading it to suggest irrelevant code. Switching back to
@codeiumdev
.
0
0
8
@MikhailBatin
@teortaxesTex
imagine if genghis khan was immortal - do you think you could still make this tweet on ur iphone?
4
0
8
so nostalgic of my photos pre YC. i’ve aged so much since then even though it’s only been a year…. need to refocus on my health and diet
2
0
7
introducing a component of our new search - semantic highlighting! it's hard to interpret why an ai thinks an opportunity is relevant.
unlike traditional keyword search where relevancy is determined largely by the amount of keywords that appear in a result, ai relevancy is
1
0
9
i'm going to call it now - a small < 7B parameter (i'm going to guess 2B) that uses Q* to achieve a higher MATH benchmark score than Gemini.
There’s something in the air. A schizo vibe of hope.
Lets get mathy.
0
49
847
1
0
6
@tobi
you'll start seeing this happen to private sector businesses as well. most companies, large and small, will start having a significant portion of their revenue coming in from government sales.
1
1
7
🚀🚀🚀
Welcome to YC,
@sachsubra
,
@PhilipKung5
, and
@SweetspotHQ
!
Sweetspot (YC S23) is building an AI-powered search engine for businesses that bid on government contracts, connecting them to all available and relevant opportunities in just seconds.
1
6
44
0
2
7
Found out today that Napoleon's last word before his death was "Josephine" - the name of his first wife. It could be inferred that a lot of his ambitions for conquering the world was solely due to impressing her. The length men would go to impress women is truly remarkable.
1
2
5
@seshubon
@inflectionAI
can you refresh the page on screen recording and show? that would be ultimate proof
1
0
4
@jxmnop
Could this be used as a heuristic to determine the quality of the embedding model? Where being able to generalize that the semantic meaning of emb(x * n) == emb(x) is a function of intelligence?
0
0
6
@aphysicist
we help you find, manage, and respond to government opportunities.
you should be able to give our search engine a natural language description of your company and it'll find you all relevant contracts - let me know if you'd like a demo!
0
0
6
for those who don't know, cursor's actual name is anysphere
1
1
6
1
0
6
0
0
6
@_Mira___Mira_
@GroqInc
they probably can - i'm sure they're just reserving a ton of capacity for 405B specifically which is why they haven't launched a paid tier yet
0
0
6
@NorthstarBrain
@elder_plinius
i don't think this is the full system prompt - theres really interesting scratchpad / internal dialogue prompts thats not included in this one.
1
0
5
we need to bring back japanese business customs - 4 being the highest acceptable rating and seppuku for being an embarrassing failure.
There's so much rating inflation on apps now
That the 0-5 rating scale is essentially 4-5 where 4 is terrible and 5 is great?
149
19
791
0
1
6
@jobergum
tantivy is a solid library with a familiar DSL - definitely was the correct design decision
0
0
6
i agree but please explain this:
5 years ago we revealed Pluribus, the first superhuman multiplayer poker AI. It cost only $150 to train. Why did poker take longer than Go? And how did it end up being so cheap? The answer is a cautionary tale on the danger of overoptimizing for benchmarks with relevance to LLMs
17
114
1K
0
0
5
Curious on everyone’s supplement stack.
I’m taking (standard dosage) of
- Tongkat Ali
- Vitamin D
- Zinc
- Fish Oil
- Lions Mane
- Magnesium
- Ashwaganda
- Multivitamin
2
1
5
@pronounced_kyle
@natosaichek
so a compressed air "rail gun"? im curious if the physics works out better for this vs magnetic acceleration.
1
0
5
such a cracked team, super happy for this public launch
Excited to more broadly share what we've been working on at
@ReworkdAI
⚡️
This past year we've fully invested in building the next generation of web agents. They're already live in production today across a handful of companies serving as the simplest way to extract web data at
32
38
260
1
0
5
@dejavucoder
fill in the middle generation, basically allowing completion models to better generate tokens between existing content. most commonly used in autocomplete codegen models but good for any sort of "autocomplete" text workflow.
1
0
4
I don't actually believe Character AI got acquired because of their inability to compete with large AI labs. It's actually the classic problem of founder market fit - Noam set out to create virtual friends to solve loneliness and instead got a bunch of underage girls hooked to
3
0
5
what's everyone's impression on posting here vs linkedin? i've been spam posting linkedin past week or so and get so much good inbound its hard to justify posting here :( (even though its more fun)
4
0
5
as an offshoot of that idea there should also be a layout.txt that contains a graph representation of each page of the site along with a summary to make it easier for browser agents to navigate
0
0
5
but why do we need to buy more bitcoin when we already own 20% of the supply?
On Saturday,
@realDonaldTrump
pledged to never sell any of the US government's bitcoin. Two days later, the Biden-Harris Administration moves $2 billion of Silk Road bitcoin. Great look and great way to reset with our industry. 🤡
498
2K
12K
2
0
4
@NickADobos
it’s also because smokers end up dying from other issues before they even develop lung cancer, which is why the statistic seems surprisingly low
0
0
4
the government spent > 40 billion on rural wifi and didn’t install a single service - you think the US government will let its most precious baby go under for 5 billion?
OpenAI is projected to lose $5 billion this year, with significant expenses on cloud computing and model training, per the Information.
131
221
2K
0
0
4
feels nice to not have to worry about a stock market crash since i liquidated all my assets to fund this startup LOL
0
0
4
1
0
4
@dustinhollywood
@Kling_ai
feels like it’s been trained on a ton of synthetic data. these outputs look reminiscent of battlefield /csgo lol
1
0
4
the omni class of models are so hard to steer im almost certain you have to fine tune them to get good performance
0
0
4
@hkarthik
amazon trained an in house model and it was complete shit so they stopped pubbing it 😂
0
0
4
I'm starting to think big tech experience is a net negative for building an early stage startup. Spent the last 3 months unlearning the "best practices" in order to ship fast :(
1
0
3